The Unix Hardware Buyer HOWTO
Eric Raymond
�����������esr@thyrsus.com
��������
Revision History
Revision 1.0 6 February 2001 Revised by: esr
Initial revision; but see the history in the introduction.
This is your one-stop resource for information about how to buy and configure
Intel hardware for cheap, powerful Unix systems.
-----------------------------------------------------------------------------
Table of Contents
1. Introduction
1.1. Purpose and History
1.2. New versions of this document
1.3. Feedback and corrections
1.4. Related resources
1.5. A discreet plug
2. Overview of the Market
3. Buying the Basics
3.1. Getting Down to Cases
3.2. Power Supplies and Fans
3.3. Motherboards
3.4. Memory
3.5. Buying a Video Card
3.6. Selecting a Monitor
3.7. Keyboards
3.8. Floppy Drives
3.9. Printers
3.10. Power Protection
3.11. Radio Frequency Interference
4. Performance Tuning
4.1. How To Pick Your Processor
4.2. Of Memory In...
4.3. Cache Flow
4.4. Suggestions for Buying
4.5. Bus Wars
4.6. Disk Wars: IDE vs. SCSI
4.7. Other Disk Decisions
4.8. Tuning Your I/O Subsystem
4.9. Souping Up X Performance
5. Hardware for Backups
6. Of Mice And Machines
7. Modems
7.1. Overview of the Modem Market
7.2. Internal vs. External
7.3. Pitfalls to Avoid
7.4. A Modem Glossary
8. CD-ROMs and Multimedia Hardware
8.1. CD-ROM Drives
8.2. Sound Cards and Speakers
9. Special considerations when buying laptops
10. How to Buy
10.1. When to Buy
10.2. Where to Buy
10.3. Computer Fairs
10.4. Mail Order
10.5. Computer Superstores
10.6. Other Buying Tips
11. Questions You Should Always Ask Your Vendor
11.1. Minimum Warranty Provisions
11.2. Documentation
11.3. A System Quality Checklist
12. Things to Check when Buying Mail-Order
12.1. Tricks and Traps in Mail-Order Warranties
12.2. Special Questions to Ask Mail-Order Vendors Before Buying
12.3. Payment Method
12.4. Which Clone Vendors to Talk To
13. Software to go with your hardware
14. Other Resources on Building Linux PCs
15. Upgrading Older Machines
15.1. Older Memory Types
15.2. If You Must Buy a 486
15.3. Cache Engineering on 486 Machines
-----------------------------------------------------------------------------
1. Introduction
1.1. Purpose and History
The purpose of this document is to give you the background information you
need to be a savvy buyer of Intel hardware for running Unix. It is aimed
especially at hackers and others with the technical skills and confidence to
go to the mail-order channel, but contains plenty of useful advice for people
buying store-front retail.
This document is maintained and periodically updated as a service to the net
by Eric S. Raymond, who began it for the very best self-interested reason
that he was in the market and didn't believe in plonking down several grand
without doing his homework first (no, I don't get paid for this, though I
have had a bunch of free software and hardware dumped on me as a result of
it!). Corrections, updates, and all pertinent information are welcomed at
[mailto:esr@snark.thyrsus.com] esr@snark.thyrsus.com. The editorial `we'
reflects the generous contributions of many savvy Internetters.
If you email me questions that address gaps in the FAQ material, you will
probably get a reply that says ``Sorry, everything I know about this topic is
in the HOWTO''. If you find out the answer to such a question, please share
it with me for the HOWTO, so everyone can benefit.
If you end up buying something based on information from this HOWTO, please
do yourself and the net a favor; make a point of telling the vendor ``The
HOWTO sent me'' or some equivalent. If we can show vendors that this HOWTO
influences a lot of purchasing decisions, we get leverage to change some
things that need changing.
Note that in December 1996 I published an introductory article on building
and tuning Linux systems summarizing much of the material in this HOWTO. It's
available here.
This document actually dates back to 1992, when it was known as the
``PC-Clone Unix Hardware Buyer's Guide''; this was before Linux took over my
world :-). Before that, portions of it were part of a Unix Buyer's Guide that
I maintained back in the 1980s on USENET.
-----------------------------------------------------------------------------
1.2. New versions of this document
New versions of the Unix Hardware Buyer HOWTO will be periodically posted to
[news:comp.os.linux.help] comp.os.linux.help and [news:
comp.os.linux.announce] news:comp.os.linux.announce and [news:news.answers]
news.answers. They will also be uploaded to various Linux WWW and FTP sites,
including the LDP home page.
You can view the latest version of this on the World Wide Web via the URL
[http://www.linuxdoc.org/HOWTO/Unix Hardware-Buyer-HOWTO/] http://
www.linuxdoc.org/HOWTO/Unix-Hardware-Buyer-HOWTO/.
-----------------------------------------------------------------------------
1.3. Feedback and corrections
If you have questions or comments about this document, please feel free to
mail Eric S. Raymond, at [mailto:esr@thyrsus.com] esr@thyrsus.com. I welcome
any suggestions or criticisms. If you find a mistake with this document,
please let me know so I can correct it in the next version. Thanks.
-----------------------------------------------------------------------------
1.4. Related resources
You may also want to look at the read the [http://www.linuxdoc.org/HOWTO/
Hardware-HOWTO.html] Hardware-HOWTO. It lists hardware known to be compatible
with Linux, and hardware known to be incompatible.
-----------------------------------------------------------------------------
1.5. A discreet plug
I finally found a vendor who consistently lives up to my standards of quality
and expertise. In fact, I liked the company so much that I accepted a seat on
their board of directors. They're [http://www.valinux.com] VA Linux Systems,
the leading vendor of pre-configured Linux machines. They're not the cheapest
(quality and careful engineering costs money) but they're the best. And
they're very active in supporting the Linux community.
-----------------------------------------------------------------------------
2. Overview of the Market
The central fact about clone hardware that conditions every aspect of buying
it is this: more than anywhere else in the industry, de-facto hardware
standards have created a commodity market with low entry barriers, lots of
competitive pressure, and volume high enough to amortize a lot of development
on the cheap.
The result is that this hardware gives you lots of bang-per-buck, and it's
getting both cheaper and better all the time. Furthermore, margins are thin
enough that vendors have to be lean, hungry, and very responsive to the
market to survive. You can take advantage of this, but it does mean that much
of the info in the rest of this document will be stale in three months and
completely obsolete in six.
One good general piece of advice is that you should avoid both the
highest-end new-technology systems (those not yet shipping in volume) and the
very cheapest systems put out by vendors competing primarily on price. The
problem with the high end is that it usually carries a hefty ``prestige''
price premium, and may be a bit less reliable on average because the
technology hasn't been through a lot of test/improve cycles. The problem with
the low end is that price-cutters sometimes settle for marginal components.
Unix works your hardware more efficiently than DOS or Windows, so it is more
sensitive to hardware flakiness, which means cut-price systems that might
deliver consistently for DOS/Windows lemmings can come around and bite you.
Use a little care, and spend the $200-$300 to stay out of the basement. The
avoided time and hassles will be worth it.
The last point deserves a little amplification. In the PC world, there's a
lot of ``if it doesn't fail, it's OK''. It is common to ignore normal
engineering tolerances (allowances for variations in components, temperature,
voltage margins, and the like) and to assume that anything which doesn't fail
outright must work. Watch out! As a historical example, the ISA bus was
originally designed for 6 MHz. IBM later updated that to 8 MHz, and that's as
much of a standard as there is, yet there were motherboards that will let you
(try to!) run it at 12 MHz, 50% over spec. Some cards were actually designed
to work at that speed with proper tolerances. Others might work...or they
might flake out when they get warm. Any systems vendor above the fly-by-night
level is going to shoot for a little more reliability than this, burning in
systems and (often) doing at least a token system test with some kind of Unix
(usually SCO XENIX). Pay the few extra bucks it costs to deal with a more
careful vendor.
The happy bottom line is this: at July 2000 direct-mail prices, you can
expect to get an AMD K6 or Pentium III 450 system with 64MB of memory, 6gig
EIDE hard disk, 3.5 floppy, 101-key keyboard, 32X CD-ROM drive, sound card &
speakers, SuperVGA-compatible 17" monitor, 56KB modem, and a decent AGP video
card for $1000 or even less. This is a more than reasonable Unix and X
machine.
I put together the first version of this guide around 1992; Unix-capable
systems are now five to ten times cheaper than they were then. At today's
prices, building your own system from parts no longer makes much sense at all
-- so this HOWTO is now more oriented towards helping you configure a whole
system from a single vendor.
-----------------------------------------------------------------------------
3. Buying the Basics
In this section, we cover things to look out for that are more or less
independent of price-performance tradeoffs, part of your minimum system for
running Unix.
Issues like your choice of disk, processor, and bus (where there is a strong
tradeoff between price and capability) are covered in the section on
performance tuning.
-----------------------------------------------------------------------------
3.1. Getting Down to Cases
Cases are just bent metal. It doesn't much matter who makes those, as long as
they're above an easy minimum quality (on some really cheap ones, cards fail
to line up nicely with the slots, drive bays don't align with the access
cutouts, or the motherboard is ill-supported and can ground out against the
chassis). If you're fussy about RFI (Radio-Frequency Interference), it's
worth finding out whether the plastic parts of the case have conductive
coating on the inside; that will cut down emissions significantly, but a few
cheap cases omit it.
Should you buy a desktop or tower case? Our advice is go with tower unless
you're building a no-expansions personal system and expect to be using the
floppies a lot. Many vendors charge nothing extra for a tower case and the
absolute maximum premium I've seen is $100. What you get for that is less
desktop clutter, more and bigger bays for expansion, and often (perhaps most
importantly) a beefed-up power-supply and fan. Putting the box and its fan
under a table is good for maybe 5db off the effective noise level, too.
Airflow is also an issue; if the peripheral bays are less cramped, you get
better cooling. Be prepared to buy extension cables for your keyboard and
monitor, though; vendors almost never include enough flex.
The airflow thing is a good argument for a full- or mid-tower rather than the
`baby tower' cases some vendors offer. However, smaller towers are getting
more attractive as boards and devices shrink and more functions migrate onto
the motherboard. A state of the art system, with all 3" disks, 200W power
supply, half-size motherboard, on-board IDE and 64meg of RAM sockets, and
half-sized expansion cards, will fit into a baby or midized tower with ample
room for expansion; and the whole thing will fit under a desk and make less
noise than a classic tower.
For users with really heavy expansibility requirements, rackmount PC cases do
exist (ask prospective vendors). Typically a rackmount case will have pretty
much the same functionality as an ordinary PC case. But, you can then buy
drive racks (complete with power supply), etc. to expand into. Also, you can
buy passive backplanes with up to 20 or so slots. You can either put a CPU
card in one of the slots, or connect it to an ordinary motherboard through
one of the slots.
-----------------------------------------------------------------------------
3.2. Power Supplies and Fans
Power supplies can matter but quality is cheap; give preference to those with
a Underwriter's Laboratories rating. There's some controversy over optimum
wattage level. On the one hand, you want enough wattage for expansion. On the
other, big supplies are noisier, and if you draw too little current for the
rating the delivered voltage can become unstable. And the expected wattage
load from peripherals is dropping steadily. The big old 300-watt supplies
that were designed for running several full-height 5.25" floppies and hard
disks are overkill in these days of portable-ready lightweight 3.5" drives.
200 watts is good enough these days, and the new breed of compact 200W
supplies is quieter to boot.
About that annoying fan noise, ask if the power-supply fan on a target system
has a variable speed motor with thermostatic control; this will cut down on
noise tremendously. However, be aware that a thermostatic sensor basically
measures the temperature at the sensor (typically within the power supply
box) and makes sure there is enough airflow to keep the power supply from
overheating. The sensor does not know a thing about the temperature in
certain hot spots likely to develop in a PC case (CPU, between SIMMs, between
drives mounted in vertically adjacent bays).
This can be a problem, because in garden variety tower cases there often
isn't enough airflow to cool all components effectively even if a single fan
is going at full speed. This is especially true if your computer has lots of
add-on cards or hard disks (not much airflow between cards or between
drives). Note that the fan in the power supply was basically designed to cool
the power supply, not the components in the case. Not providing additional
fans is a sign of cheapness. On tower PCs with "expensive" engineering (e.g.
HP Vectra, Compaq) one will find one to two extra fans besides the one in the
power supply.
So the bottom line is, use thermostatic controls if you can to cut noise. But
if you want high reliability, use two or more fans. Modern designs will
usually have a small auxilliary fan mounted right over the chip! This is a
good feature to ask about in advance.
The noise produced by a fan is not just a function of the speed with which it
turns. It also depends on the nature of the airflow produced by the fan
blades and the bearings of the rotor. If the blades causes lots of turbulent
airflow, the fan produces lots of noise. One brand of fans that, I'm told,
has a reputation for being much more silent than others even if going at full
throttle is the German manufacturer Pabst. Their fans are offered in US
computer magazines.
-----------------------------------------------------------------------------
3.3. Motherboards
Provided you exercise a little prudence and stay out of the price basement,
motherboards and BIOS chips don't vary much in quality. There are only six or
so major brands of motherboard inside all those cases and they're pretty much
interchangeable; brand premiums are low to nonexistent and cost is strictly
tied to maximum speed and bus type. Unless you're buying from a "name" outfit
like Compaq, Dell, or AST that rolls its own motherboards and BIOSes, there
are only four major brands of BIOS chip (AMI, Phoenix, Mylex, Award) and not
much to choose between 'em but the look of the self-test screens. One
advantage Unix buyers have is that Unixes are built not to rely on the BIOS
code (because it can't be used in protected mode without more pain than than
it's worth). If your BIOS will boot properly, you're usually going to be OK.
The ``Baby AT'' motherboard form factor is obsolete. It's an ATX world now.
There are still a few potential gotchas to beware of, especially in the
cheaper off-brand boards. One is ``shadow RAM'', a trick some boards use for
speeding up DOS by copying the ROM contents into RAM at startup. It should be
possible to disable this. Also, on a cacheing motherboard, you need to be
able to disable cacheing in the memory areas used by expansion cards. Some
cheap motherboards fail to pass bus-mastering tests and so are useless for
use with a good SCSI interface; on others, the bus gets flaky when its turbo
(high-speed) mode is on. Fortunately, these problems aren't common and are
becoming less so.
You can avoid both dangerously fossilized hardware and these little gotchas
by sticking with a system or motherboard design that's been tested with Unix.
Some other good features to look for in a motherboard include:
��*�Gold-plated contacts in the expansion slots and RAM sockets. Base-metal
contacts tend to grow an oxidation layer which can cause intermittent
connection faults that look like bad RAM chips or boards. (This is why,
if your hardware starts flaking out, one of the first things to do is
jiggle or remove the boards and reseat them, and press down on the RAM
chips to reseat them as well -- this may break up the oxidation layer. If
this doesn't work, rubbing what contacts you can reach with a soft eraser
is a good fast way to remove the oxidation film. Beware, some hard
erasers, including many pencil erasers, can strip off the plating, too!)
��*�The board should be speed-rated as high as your processor, of course.
It's good if it's rated higher, so upgrade to a faster processor is just
a matter of dropping in the chip and a new crystal.
If you're changing a motherboard, see the [http://www.atipa.com/InfoSheets/
instmb.shtml] Installing a Motherboard page first. This one even has a Linux
note.
-----------------------------------------------------------------------------
3.4. Memory
All current PC designs include a cacheing memory controller and some SRAM
that combine to produce higher effective speeds. Be sure to read the `Cache
Flow' section under performance tuning for more on how cache configuration
affects your performance.
Current motherboards use PC100 SDRAM, which comes packaged on 168 pin DIMM
modules. SDRAM is a large step forward in memory speed, at 10ns. SDRAM does
not need to be installed in pairs. The key words you want to see on the spec
sheet are:
��*�SDRAM DIMM -- the physical module type.
��*�ECC -- error correcting memory, important for reliability; Unix makes
more efficient use of hardware and thus beats on your memory harder than
Windows does.
��*�PC100 -- this is the memory-module specification for current motherboards
with a 100Mhz PCI bus.
��*�16x72 -- if it says 16x64, the extra 8 bits per 64 needed for ECC are not
present.
Also, avoid anything that says `ECC compatible memory'; this is the opposite
of `memory with ECC'.
For more technical stuff on memory architectures, see The Ultimate Memory
Guide maintained by Kingston Technologies.
-----------------------------------------------------------------------------
3.5. Buying a Video Card
Video controllers translate byte values deposited in their video memory by
your GUI (usually an X server under Linux) into an analog RGB signal which
drives your monitor. The simplest kinds treat their video memory as one big
frame buffer, requiring the CPU to do all dot-painting. More sophisticated
``accelerated'' cards offer operations such as BitBlt so your X server can
hack the video memory algorithmically. These days almost all cards even at
the low end actually have some acceleration features.
Cards are rated by the maximum number of analog signal changes they can
produce per second (video bandwidth). Video bandwidth can be used to buy
varying combinations of screen resolution and refresh speed, depending on
your monitor's capabilities.
Another important variable of video cards is the size of their on-board video
RAM. Increased memory lets you run more colors at higher resolutions. For
instance, a 1MB card usually will only allow 256 colors at 1024x768 while a
2MB card usually allows at least 16-bit color (a palette of about 65,000
colors). You'll need 4MB of video memory to use 24-bit or ``true'' color (16
million colors) at 1024x768.
The card's video RAM size has no effect on its speed. What does affect speed
is the type of memory on board. VRAM (Video Random Access Memory) is fast but
more expensive; it features a dual-ported design allowing two devices (the
CRT controller and the CPU) to access the memory at the same time. DRAM
(Dynamic Random Access Memory) is is similar to the RAM used in main
memories. It is cheaper, more common, and slower (because the CRT controller
and the CPU must take turns accessing the video buffer).
A quick review of monitor standards:
Table 1. Monitor standards
+------------+----------+------+------------------+----------------+--------+
|Name |Resolution|Colors|Horizontal |Vertical |Notes |
| | | |Frequency |Frequency | |
+------------+----------+------+------------------+----------------+--------+
|MDA |720x350 |� |18.43 KHz |50 Hz |Obsolete|
+------------+----------+------+------------------+----------------+--------+
|CGA |640x200 |2 |15.85 KHz |60 Hz |Obsolete|
| +----------+------+ | | |
| |320x200 |4 | | | |
+------------+----------+------+------------------+----------------+--------+
|EGA |640x350 |16 |21.80 KHz |60 Hz |Obsolete|
+------------+----------+------+------------------+----------------+--------+
|VGA |640x480 |16 |31.50 KHz |60 Hz |� |
| +----------+------+ | | |
| |320x200 |256 | | | |
+------------+----------+------+------------------+----------------+--------+
|VESA VGA |640x480 |16 |38.86 KHz |72 Hz |� |
| +----------+------+ | | |
| |320x200 |256 | | | |
+------------+----------+------+------------------+----------------+--------+
|VESA SVGA |800x600 |16 |48.01 KHz |72 Hz |� |
| +----------+------+ | | |
| |640x480 |256 | | | |
+------------+----------+------+------------------+----------------+--------+
|8514/A |1024x768 |16 |35.20 KHz |43.5 Hz |Obsolete|
+------------+----------+------+------------------+----------------+--------+
|XGA |64x480 |65536 |?? |?? |Obsolete|
| +----------+------+ | | |
| |1024x768 |256 | | | |
+------------+----------+------+------------------+----------------+--------+
|VESA |1024x768 |256 |56.48 KHz |70 Hz |� |
|1024x768 | | | | | |
+------------+----------+------+------------------+----------------+--------+
The Horizontal and Vertical Frequency columns refer to the monitor scan
frequencies.
The vertical frequency is the upper limit of the monitor's flicker rate; 60Hz
is minimal for ergonomic comfort, 72Hz is VESA-recommended, and 80Hz is
cutting-edge. At resolutions above VGA, horizontal scans take long enough
that the monitor may never reach anywhere near the vertical-frequency
maximum; how close it gets is a function of the horizontal-scan frequency
(higher is better).
For more information on how to avoid the evil screen flicker, see the XFree86
Video Timings HOWTO, a tutorial written by your humble editor and included
with the XFree86 distribution.
It's still barely possible to find MDA video boards and monitors out there,
but prices for SVGA have collapsed so totally that it's not worth bothering.
XGA is an IBM-proprietary included for completeness, but is vanishingly rare
in the clone market. 8514/A is another IBM standard supported by a few
graphics accelerator cards. It is interlaced, and thus has a tendency to
flicker. The VESA 1024x768 standard makes XGA and 8514/A obsolete.
SVGA or `Super VGA' strictly refers only to 800x600 resolution, but is widely
used for 1024x768 and even 1280x1024 resolutions. Standards above 1024x768
are weak and somewhat confused.
These days, most vendors bundle a 14" monitor and super-VGA card with
1024x768 resolution in with their systems. Details to watch are the amount of
RAM included (which will affect how much of that maximum resolution and how
many colors you actually get), and whether the memory is dual-ported VRAM
(slightly more expensive but much faster).
You should check ahead of time which Super VGA chipset the vendor normally
ships. Though DOS/Windows doesn't really care, the Unix software that uses it
(most notably X servers) will definitely notice the difference. Many
commercial implementations of X servers for Intel Unix don't know how to use
the SVGA modes of the cheapie Oak and Trident SVGA chipsets, which are the
ones most often bundled with systems (XFree86 handles them OK, though). The
cards based on the Tseng 4000 chipsets are often bundled or available as as
an extra-cost option (usually less than $50 above Oak or Trident), perform
better, and are supported by the most implementations of Intel Unix-based X
servers. ATI VGAWonder chipset is well-supported and often bundled with
multimedia or graphics systems.
More information on video standards is available at the Video Display
Standards page of the PC Guide site.
-----------------------------------------------------------------------------
3.6. Selecting a Monitor
3.6.1. What To Look For On The Spec Sheet.
��*�Dot pitch of 0.28 or smaller on a 12"-15" monitor; 0.30 is acceptable on
larger ones, especially 19" to 21" screens (but look extra hard at 0.25
21-inchers like the Viewsonic 21PS or Nokia 445X). Dot pitch is the
physical resolution of the screen's phosphor mask. Larger dot pitches
mean that small fonts and graphic details will be fuzzy.
��*�72Hz or better vertical scan frequency, to cut flicker.
��*�Non-interlaced display. Interlacing cuts the required scan frequency for
a given resolution in half, but makes flicker twice as bad. As a result,
interlaced monitors are rapidly disappearing; don't get stuck with one.
��*�Multisync capability. These days, only the cheapest low-end monitors have
fixed horizontal-scan frequencies. If you buy one, it's potluck whether
it will work with your present graphics board, let alone your next one.
Your monitor is a big investment, and likely to outlive a generation or
two of graphics boards. Don't get stuck!
��*�Does it have a tilt-and-swivel base? Adequate controls, including both
horizontal and vertical size and horizontal and vertical centering? A
linearity control, a trapezoidal control, and a color-temperature control
are all pluses; the last is particularly important if you compose
graphics on screen for hardcopy from a printer.
��*�Is it color? If you don't see it in the ad, ask; some lowball outfits
will try to palm off so-called "black & white VGA" monitors on you.
��*�For X use, a 14", .28mm dot pit, non-interlaced 72Mhz monitor at 640x480
resolution is the bare minimum for comfortable use, and that resolution
leaves you rather squeezed for screen real estate. 1024x768 is much
better. If your budget will stand it at all, a 17" or 20" monitor is a
good investment. A 17" monitor is minimum if you're going to go with
1280x1024 resolution.
If you can, buy your monitor from someplace that will let you see the same
monitor (the very unit you will walk out the door with, not a different or
`demo' unit of the same model) that will be on your system. There's a lot of
quality variation (even in "premium" monitor brands) even among monitors of
the same make and model.
Another good reason to see before you buy, and carry it home yourself, is
that a lot of monitors are vulnerable to bumps. The yoke can get twisted,
producing a disconcerting tilt in the screen image.
The Caveat Emptor guide has a good section on evaluating monitor
specifications. And there's a database of monitor specs at [http://
www.interlog.com/~gscott/TBOML/MAIN.html] The Big Old Monitor List.
-----------------------------------------------------------------------------
3.6.2. Eric Buys A Big Monitor: Smart Shopping Tips
In early 1996 the good folks at O'Reilly Associates dropped several $1000
checks on me in relatively quick succession (payment for fast-turnaround
technical reviews). I decided to use the money to treat myself to a really
good monitor.
This page tells you how I did it. Specific specs and pricing information will
date quickly, but the method should still be good years from now.
My existing monitor wasn't bad -- a 17-inch Swan 617 that I could drive at a
bit above 1024x768. Still, I yearned for more real estate -- especially
vertical real estate, so I could view full PostScript pages using a legible
font.
This brings us to our first prescription: be clear about what you want. It's
easy, and very expensive, to buy more monitor than you'll really use.
I knew I wanted something in the 19-to-21-inch range, with 1280x1024 or
higher resolution. I knew this would probably cost me about $2000, and could
afford it. I knew I didn't need one of the monster projection monitors
further upmarket, with screen sizes 24" and up. These will typically cost you
$4K or so and are too big for desktop use anyway.
I also knew I didn't need one of the special true-color monitors designed for
photo composition, making print separations, and so forth. These creatures
(always Trinitrons) have better, denser color than conventional tubes but at
a hefty price premium (and usually at some cost in available resolution). If
all you're going to do most of the time is 16 or 256-color X screens, you
don't need this capability.
Once you've settled on what you need, gather comparative data. It was 1996,
so I started out by making phone calls to manufacturer 800 numbers. Then I
discovered that almost all the manufacturers had Web sites, with technical
specs for their monitors on them. Today, you'd go to the Web first.
(This space used to inclide detailed technical data on what I found " model
numbers, resolutions, reviews of manufacturer websites, etc." but I've
removed it because it's all five years out of data now.)
This wasn't at all a hard call. The ViewSonic 21PS and Nokia 445X stood out
from the pack immediately; their combination of high bandwidth with a 21-inch
screen size and ultra-fine .25 dot pitch promised better performance than the
general run of .28-pitch monitors.
Nor was the choice between the two very hard. ViewSonic's 21PS is $600 less
expensive than Nokia's 445X for very similar performance. And, other things
being equal, I'd rather buy a monitor from a specialist monitor manufacturer
than a general consumer electronics outfit best known for its cellular
phones.
So I determined to order a ViewSonic 21PS.
This left me with a second problem. My ATI Mach 32 can't drive a monitor at
higher than 1280x1024 resolution and 94MHz bandwidth. So it wouldn't be able
to drive the 21PS at 1600x1200. I wound up buying a Mach 64.
The combination worked wonderfully (two years later I discovered that [http:/
/www.valinux.com] VA Linux Systems buys the same monitor for its high-end
systems). The only problem I have with it is that monitor is way bright even
dialed down to its dimmest setting. You'll need a strong light in the room
where you install it. Also, be aware that the only really convenient way to
move one of these monster monitors is with a forklift!
-----------------------------------------------------------------------------
3.7. Keyboards
It's important to get a high quality keyboard with good key feel. See the
typing-injury FAQ from [news:sci.med.occupational] sci.med.occupational to
see what happens if you don't. Carpal tunnel syndrome is no fun for anyone,
but it hits hackers particularly hard. Don't be a victim!
Hal Snyder of Mark Williams, Co sent us the following caveat: "We find that
about 10% of cheap no-name keyboards do not work in scan code set 3. We are
interested in scan code set 3 because only there can you reprogram the
keyboard on a per-key basis as to whether keys are make-only, make-break, or
autorepeat. It is a big win for international support and for X."
He continues: "Keytronic, Cherry, and Honeywell keyboards, as well as a large
number of imports, work fine. My advice is to either buy a respected brand of
keyboard, or deal with a vendor who will allow you to return an incompatible
keyboard without charge."
Allen Heim <ajh@netscape.com> writes: I'm sold on NMB keyboards ([http://
www.nmbtech.com/] http://www.nmbtech.com/), available from Global Computer
([http://www.globalcomputer.com/] http://www.globalcomputer.com/), or at
(800) 8-GLOBAL. Their line of mechanical-switch keyboards now have a lifetime
warranty, and I've just ordered my second RT-8200 unit. I don't see it listed
on their website currently, but I do see their "Windows 95-enhanced" model,
the RT-8200W. It's the same thing, but with extra keys (which may be
programmable; think of the Emacs "meta" key--could be useful).
In the wake of recent publicity about reprtitive-strain injuries, ergonomic
keyboards are becoming increasingly common. One that looks promising to your
editor (though I haven't yet used it) is the Marquardt MiniErgo MF2, from
Marquardt Switches, Inc.; 2711 Route 20 East, Cazanovia NY 13035, phone (315)
-655-8050; suggested list price $170, AT-compatible interface). Michael Scott
Shappe <mshappe@jeeves.net> sent me a rave review of the Marquardt after
having used it about six months.
The MF2 features a conventional QWERTY layout, but with the right and left
halves split apart and rotated about 30 degrees towards each other in a
shallow V shape. The theory is that being able to angle your arms inward and
your elbows out produces a less stressful typing position.
The MF2 has no keypad, but it does have the standard 12 function keys across
the top and arrow keys at the point of the V (meant to be thumb-operated).
I have seen and used a device called the Maltron ergonomic keyboard. This
keyboard splits the keys into two main groups, each arranged in a dished
hollow. Each hand also has easy access to separate thumb pads of nine keys
each; there is one numeric and arrow-key pad in the center of the unit,
between the right and left-hand groups. Also, the keys in each group do *not*
have the alternate-row staggering of the conventional (Scholes) keyboard;
this subtle change reduces torsional stress on the fingers and wrists
tremendously.
I found the Maltron easy to use, and regretted having to give it back.
For more details on many ergonomic keyboards and typing-injury issues in
general, see Dan Wallach's FAQ on repetitive strain injuries and ergonomic
input devices, published monthly in [news:news.answers] news.answers
-----------------------------------------------------------------------------
3.8. Floppy Drives
There's not much to be said about floppy drives. They're cheap, they're
generic, and the rise of CD-ROM drives as a cheap distribution medium has
made them much less important than formerly.
The media come in two sizes -- 3.5-inch `hard-shell' floppies and 5.25"
flexible floppies. The 3.5" variety carries 1.44MB of information; the older,
larger 5.25" carries 1.2MB. The 5.25" variety is obsolete, but is still
included with some new systems as a sort of vermiform appendix, like the
useless ``Turbo'' buttons on many cases.
You'll probably never use floppies for anything but first boot of a new
operating system, and can almost certainly do without a 5.25" drive entirely.
Go ahead and settle for cheap Mitsumi and Teac floppy drives. There are no
`premium' floppy drives anymore. Nobody bothers. If you're seriously short of
expansion-bay space, you can get a combination drive with 5.25 and 3.5" slots
on a single spindle.
-----------------------------------------------------------------------------
3.9. Printers
There's one huge gotcha about printers: Don't Get a `GDI' printer!
-----------------------------------------------------------------------------
3.9.1. GDI printers: avoid!
Low-end printer manufacturers have been increasingly moving towards `GDI' or
`Windows' printers for their laser or LED models. These have the unfortunate
characteristic of being, at worst, unusable from Unix; and at best, useable
only with reduced resolution (e.g., 300 dpi from a 600 dpi printer).
The problem is that the design of laser printers inherently requires that
data move onto the imaging drum at a precisely-controlled rate, and so laser
(and LED) printers have traditionally included a CPU of moderate speed and
enough RAM to image an entire page, either as a complete bitmap or as ASCII,
using a quick font rasterizer to form images on-the-fly. GDI printers,
however, offload this responsibility to the computer, and therefore require
very specialized drivers that (a) are not available for any Unix, AFAIK; and
(b) slow down the computer a lot when printing is underway.
Most GDI printers DO support HP's PCL in a lower resolution, so they can
often be used from Unix via Ghostscript, but only in 300 dpi and/or missing
some features. They may also be slower in this mode than in their native GDI
modes. In order to print at 600 dpi with Ghostscript, a PCL printer must
support HP's PCL at level 5e or better, so printer purchasers should look for
that in any non-PostScript model, at least at the moment. (Naturally, all of
this could change if/when support is added to Ghostscript for more esoteric
models; but AFAIK, this is the current set of limitations.)
Finally, I've seen one extra twist on this already-twisted marketplace: The
Brother HL-720 is advertised as supporting `PCL 5e for DOS' (or words to that
effect). What this means is that it's a GDI model with a DOS driver that
takes PCL 5e input and translates it to the printer's native GDI mode.
Needless to say, this is useless for Unix.
GDI printers are a bad design even for the DOS lemmings, because they slow
the machine down significantly while printing is going on. Like `WinModems',
they're a sleazy way for manufacturers to save a few bucks. Our advice is,
buy a printer with native Postscript and avoid all this crap.
-----------------------------------------------------------------------------
3.9.2. Non-GDI printers
There really isn't all that much to be said about printers; the market is
thoroughly commoditized and printer capabilities pretty much independent of
the rest of your hardware. The PC-clone magazines will tell you what you need
to know about print quality, speed, features, etc. The business users they
feed on are obsessed with all these things.
Most popular printers are supported by GhostScript, and so it's easy to make
them do PostScript. If you're buying any letter-quality printer (laser or
ink-jet), check to see if it's on GhostScript's supported device list --
otherwise you'll have to pay a premium for Postscript capability! Postscript
is still high-end in the MS-DOS market, but it's ubiquitous in the Unix
world.
Warning, however: if you're using ghostscript on a non-Postscript printer,
printspeed will be slow, especially with a serial printer. A bitmapped 600
dpi page has a lot of pixels on it. Further, if you're doing much printing,
ghostscript will create enormous spool files. (megabytes/page). At today's
prices, paying the $750 or so for Postscript capability makes sense.
If you're buying a printer for home, an inkjet is a good choice because it
doesn't use gobs of power and you won't have the toner/ozone/noise/etc mess
that you do with a laser. If all you want is plain-ASCII, dot-matrix is
cheaper to buy and run.
Inkjets are great in that they're cheap, many of them do color, and there are
many kinds which aren't PCL but are understood by Ghostscript anyway. If you
print very infrequently (less than weekly, say), you should be careful to buy
a printer whose print head gets replaced with every ink cartrige: infrequent
use can lead to the drying of the ink, both in the ink cartrige and in the
print head. The print heads you don't replace with the cartrige tend to cost
nearly as much as the printer (~$200 for an Epson Stylus 800) once the
warranty runs out (the third such repair, just after the warranty expired,
totalled one informant's Stylus 800). Be careful, check print head
replacement costs ahead of time, and run at least a cleaning cycle if you
don't actually print anything in a given week. (Conversely, toner starts out
dry, and ribbon ink won't evaporate for years...if you truly print only
rarely, but neither a dot matrix nor a laser makes sense, consider buying no
printer and taking your PostScript files to a copy shop...)
A parallel interface is a cheap way to make your printer print a lot faster
than a serial line, and everyone's got a parallel port in their PC.
A few printers for the MS-DOS market require a special controller card and
proprietary cable to do PostScript. These require MS-DOS software and
typically won't run under Unix at all.
Meanwhile, there are several true 600 dpi lasers that grok PCL 5e, yet cost
less than $500 retail. Currently (December 1997) these include the Lexmark
Optra E (and E+), the HP 5L (and 5L with suffix, and probably 6L), and the
Brother 760. As you can't easily buy a new hard drive smaller than 2
gigabytes, tens of megabytes of spare space in /var/spool should be the
accepted norm, rather than a problem, for new systems; I've also noticed that
PCL 5e seems to include some amount of compression (probably RLE or font
encoding) which works rather well for text, further reducing the spool
requirements.
One of our spies says good things about the Canon BJC-240 and 250. He reports
they preint well with Ghostscript and are more reliable than Deskjets.
I personally have a LaserJet 6MP, and like it.
-----------------------------------------------------------------------------
3.10. Power Protection
3.10.1. Overview
Finally, I strongly recommend that you buy a power conditioner to protect
your hardware. MOV-filtered power bars make nice fuses (they're cheap to
replace), but they're not enough.
The technical info in the remainder of this section is edited from material
supplied by David E. Wexelblat >dwex@mtgzfs3.att.com<.
There are several levels of power protection available to the home computer
user. I break this down into 4 levels; others may have different ways of
classifying things. The levels are:
1. Surge Suppressor
2. Line Conditioners
3. Standby Power Supplies
4. Uninterruptible Power Supplies
And here's what they mean:
-----------------------------------------------------------------------------
3.10.2. Surge suppressors
These are basically a fancy fuse between the source and your hardware; they
clamp down spikes, but can't fill in a low voltage level or dropout.
This is a bare minimum level of protection that any piece of expensive
electronics should have. Note that this applies to more than just AC power;
surge suppressors are available for (and should be used on) phone lines, and
RS-232 and parallel connections (for use on long lines; generally not needed
if the devices is colocated with the computer and all devices are protected
from outside sources). Note also that all devices connected to your computer
need to be protected; if you put a surge suppressor on your computer but not
your printer, then a zap on the printer may take out the computer, too.
An important fact about surge suppressors is that they need to be replaced if
they absorb a large surge. Besides fuses, most suppressors rely on on
components called Metal-Oxide Varistors (or MOVs) for spike suppression,
which degrade when they take a voltage hit. The problem with cheap
suppressors is that they don't tell you when the MOV is cooked, so you can
end up with no spike protection and a false sense of security. Better ones
have an indicator.
You can buy surge suppressors at any Radio Shack; for better prices, go
mail-order through Computer Shopper or some similar magazine. All of these
are low-cost devices ($10-50).
-----------------------------------------------------------------------------
3.10.3. Line Conditioners
These devices filter noise out of AC lines. Noise can degrade your power
supply and cause it to fail prematurely. They also protect against short
voltage dropouts and include surge suppression.
My Tripp-Lite 1200 was typical of the better class of line conditioners; a
box with a good big soft-iron transformer and a couple of moby capacitors in
it and no conductive path between the in and out sides. With one of these,
you can laugh at brownouts and electrical storms. You can get for $139 or so
by mail order. A fringe benefit of this little beauty is that if you
accidentally pull your plug out of the wall you may find you actually have
time to re-connect it before the machine notices (I did this once). But a
true SPS or UPS is better.
Netter Trey McLendon has good things to say about Zero Surge conditioners. He
says: "Our systems at work [...] have been protected for 2.5 years now
through many a violent storm...one strike knocked [out] the MOV-type
suppressors on a Mac dealer's training setup across the street from us. The
Zero Surge just sort of buzzed when the surge came in, with no interruption
whatsoever. The basic principle is this: ZS units slow down the surge with a
network of passive elements and then sends it back out the neutral line,
which is tied to ground _outside at the box_ by code. MOV units shunt the
surge to ground _at the computer_, where it leaps across serial ports,
network connections, etc. doing its deadly work."
Price vary widely, from $40-400, depending on the power rating and
capabilities of the device. Mail-order from a reputable supply house is your
best bet. Line conditioners typically don't need to be replaced after a
surge; check to see if yours includes MOVs.
-----------------------------------------------------------------------------
3.10.4. Standby power supplies (SPSs)
These devices are battery-based emergency power supplies that provide power
for your system via an inverter if the power fails. An SPS will generally
have all the capabilities of a line conditioner as well.
Note: these devices do not come on line until after the power fails, and have
a certain amount of delay (typically some milliseconds) before they come on
line. If the capacitors in your power supply are not large enough, the SPS
may not cut in soon enough to prevent your computer from seeing the power
failure.
Note also that many SPSs are marketed as Uninterruptable Power Supplies (see
below). This is incorrect. Any device with a non-zero cutover time cannot be
a true UPS. If the ad mentions a cutover time, it's an SPS, and not a UPS.
The price range for these devices (increasing with increasing peak load
capacity and with decreasing cutover time) is $200-2000. An SPS will not need
to be replaced after absorbing a large surge.
-----------------------------------------------------------------------------
3.10.5. Uninterruptable power supplies (UPSs)
These devices provide full-time isolation from the incoming AC line through a
transformer of some sort. These devices are on-line at all times, and if the
AC line fails, the batteries will cut in. Your devices will see no
interruption of their incoming AC. UPSs cost more, and provide more features.
They are the ultimate in power protection. Many UPSs have an intelligent
interface that will notify a connected device of a power failure, allowing it
to shut down cleanly. UPSs also provide the capabilities of a line
conditioner. The price range (for devices in the size range for a home
computer) are $200-$1500. An UPS will not need to be replaced after absorbing
a large surge.
Now, given this information, how does one decide what to get? For a system
that runs unattended, like most Unix systems, it is best to have a device
that provides both power holdover and a power failure signal. Hence, for a
Unix system, a UPS or SPS with Unix monitoring software is the best choice.
If the vendor isn't secretive about interface specs, it's fairly simple to
write your own daemon to monitor a serial port, and send init a SIGPWR signal
when it sees a powerdown notification on the port. Freeware power-monitor
demons are available for Linux.
Many UPS/SPS signal ports work by asserting a pin, so that one could use a
modem-control serial port on the PC and wire this pin to "Carrier Detect" in
order to monitor it. Some, like the APC "SmartUPS" series, actually conduct
an ASCII dialog with the host through a serial line in order to accomplish
the monitor functions.
Our recommendation for a production Unix environment is a configuration like
the following:
1. An on-line UPS or SPS for the computer system. An intelligent interface
is mandatory, along with appropriate software for ordered shutdown.
2. Surge suppression on all phone lines, and also on serial/parallel lines
that leave the room.
3. Line conditioners on any devices not connected to the UPS. If you do take
a power hit, it's cheaper to replace a $50 line conditioner than a $1500
laser printer.
If this is too expensive for you, then downgrade the UPS/SPS to a line
conditioner like the TrippLite. But don't go without at least that. Running
unprotected is false economy, because you will lose equipment to electrical
storms ?? and, Murphy's Law being what it is, you will always get hit at the
worst possible time.
An important question is "How do I know how big a UPS/SPS to get?" The watt
rating of the UPS/SPS should be at least the sum of the peak ratings off all
equipment connected to it (don't forget the console monitor). Power-supply
marketroids tend to quote you capacities and formulas like "sum of VA ratings
+ 20%" which (surprise!) push you towards costlier hardware. Ignore them. If
a watt rating is not given, watts = 0.75*VAmax.
One other consideration is that you typically shouldn't put a laser printer
on a UPS ?? toner heaters draw enough current to overload a UPS and cause a
shutdown within seconds. The other thing is that you can't even put the laser
printer on the same circuit with a UPS ?? the heater kicks on every 20-30
seconds, and most UPSs will see the current draw as a brownout. So buy a
separate line conditioner for the laser printer.
Finally, read the UPS's installation manual carefully if you're going to use
it with other power-protection devices. Some UPSs don't like having surge
suppressors between them and the equipment.
David personally recommends surge suppressors and line conditioners from
Tripp-Lite (available both mail-order and retail), and UPSs from Best Power
Technologies (Necedah, WI - 1-(800)-356-5737). I can enthusiastically second
the TrippLite recommendation, but haven't dealt with Best Power at all.
Tripp-Lite has a whole range of products, from a $10 phone-line
surge-suppressor, to line conditioners and SPSs with prces in the hundreds of
dollars. They have a line of $50-80 line conditioners that are good for most
peripherals (including your home stereo :->).
Best Power Technologies sells two lines of UPSs in the range for home
systems. The older and more expensive FERRUPS line (which is what David has)
has a smart interface, and very good filtering and surge-suppression
capabilities. He says "I have a 1.15kVA FERRUPS for my home system, which is
overkill with my current hardware (although it rode out a 45 minute power
failure with nary a whisper - no reboot). In 1990, I paid ~$1600 for this
device, and that has since gone up. They also sell a newer line of Fortress
UPSs. These are better suited in price for home systems. I don't know much
about them, as they were not available when I bought my UPS. I expect that
this is what most people will want to consider, though. In addition, Best
sells Check-UPS, a software package (in source form) for monitoring the UPS
and shutting it down. I have found Best to be a good company to deal with,
with competent, knowledgeable sales people (who will be able to help you pick
the right device), and helpful, courteous, and responsive technical support."
Other things to know:
A UPS should be wired directly to (or plugged directly into) the AC supply
(i.e. a surge suppressor is neither required nor suggested between the wall
and the UPS). In addition, a surge suppressor between the UPS and the
equipment connected to it is redundant.
-----------------------------------------------------------------------------
3.11. Radio Frequency Interference
(Thanks to Robert Corbett <Robert.Corbett@Eng.Sun.COM> for contributing much
of this section)
Radio Frequency Interference (RFI) is a growing problem with PC-class
machines. Today's processor speeds (above 33MHz) are such that the
electromagnetic noise generated by a PC's circuitry in normal operation can
degrade or jam radio and TV reception in the neighborhood. Such noise is
called Radio Frequency Interference (RFI). Computers, as transmitting
devices, are regulated by the Federal Communications Commission (FCC).
FCC regulations recognize two classes of computer:
If a PC is to be used in a home or apartment, it must be certified to be FCC
class B. If it is not, neighbors have a legal right to prevent its use. FCC
class A equipment is allowed in industrial environments.
Many systems are not FCC class B. Some manufacturers build boxes that are
class B and then ship them with class A monitors or external disk drives.
Even the cables can be a source of RFI.
It pays to be cautious. For example, the Mag MX17F is FCC class B. There are
less expensive versions of the MX17 that are not. The Mag MX17 is a great
monitor (I wish I had one). It would be painful to own one and not be allowed
to use it.
An upgradeable system poses special problems. A system that is FCC class B
with a 33 MHz CPU might not be when the CPU is upgraded to a 50 or 66 MHz
CPU. Some upgrades require knockouts in the case to be removed. If a knockout
is larger than whatever replaces it, RFI can leak out through the gap.
Grounded metal shims can eliminate the leaks.
Even Class B systems don't mix will with wireless phonesets (not cellular
phones, but the kind with a base station and antennaed headset). You'll often
find a wireless phone hard to use withing 20 feet of a Class B machine.
To cut down on RFI, get a good metal case with tight joints, or at least make
sure any plastic one you buy has a conductive lining. You can also strip the
painted metal-to-metal contacting parts of paint so that there's good
conductive metal contact. Paint's a poor conductor in most cases, so you can
get some benefit from this.
-----------------------------------------------------------------------------
4. Performance Tuning
Here are the places where you can trade off spending against the performance
level you want to buy and your expected job mix.
-----------------------------------------------------------------------------
4.1. How To Pick Your Processor
Right now, the chips to consider for running Unix are the Pentium IIs and
Pentium IIIs and their clone equivalents from AMD or Cyrix. Life used to be
more complicated, but with Pentium prices plunging as they have been and the
PCI bus having taken over the world, nothing else makes much sense for a new
Intel-based system.
Brands don't matter much, so don't feel you need to pay Intel's premiums if
you see an attractive Cyrix, AMD or other chip-clone system offered.
To compare the performance of different Intel-based systems with each other
and with machines from other manufacturers, you can take a look at the
SPECmark Table at [ftp://ftp.cdf.toronto.edu/pub/spectable] ftp://
ftp.cdf.toronto.edu/pub/spectable. That document recommends (and I do too)
that you read the SPEC FAQ at [http://www.specbench.org/spec/faq] http://
www.specbench.org/spec/specfaq.html to get background before browsing the
table.
Good current advice about chipsets can be found at [http://
www.math.sunysb.edu/~comech/tools/CheapBox.html#mobo] The Cheap /Linux/ Box.
-----------------------------------------------------------------------------
4.2. Of Memory In...
Buy lots of RAM, it's the cheapest way to improve real performance on any
virtual-memory system. 64MB now comes standard on most clone configurations.
This is good enough for X.
Tuning is simple. Watch your job mix with top(1) and add memory until you're
not swapping to disk any more.
-----------------------------------------------------------------------------
4.3. Cache Flow
4.3.1. Overview
The most obscure of the important factors in the performance of a clone
system is the motherboard's memory subsystem design, consisting both of
primary and secondary cache and DRAM. The two questions performance-minded
buyers have to deal with are: (1) does the cache design of a given
motherboard work with Unix, and (2) how much cache SRAM should my system
have, and how should it be organized?
Before normal clock speeds hit two digits in MHz, cache design wasn't a big
issue. But DRAM's memory-cycle times just aren't fast enough to keep up with
today's processors. Thus, your machine's memory controller caches memory
references in faster static RAM (SRAM), reading from main memory in chunks
that the board designer hopes will be large enough to keep the CPU
continuously fed under a typical job load. If the cache system fails to work,
the processor will be slowed down to less than the memory's real access speed
?? which, given typical 70ns DRAM parts, is about 7MHz.
You'll sometimes hear the terms L1, L2, and L3 cache. These refer to Level 1,
Level 2, and Level 3 cache. L1, today, is always on the CPU (well, unless
you're HP). L2 is off-chip cache. L3 is a second-level off-chip cache.
Anything that will fit in L1 can be run at full CPU speed, as there is no
need to go off chip. Anything (or things) too large to fit in L1 will try to
run in L2. If it fits in L2, you still won't have to deal with the bus. If
you're familiar with virtual memory, think of it this way: When you run out
of L1, you swap to L2, when you run out of L2, you swap to main memory, when
you run out of main memory, you swap to disk. Each stage is slower, and more
prone to conflicts with other parts of the system, than what it follows.
Cache is like memory, but is faster, more expensive, and smaller. L1 is
generally faster and smaller than L2, which is generally faster and smaller
than L3, which is faster and smaller than memory. Some PCs will not have L2
or L3. Most workstation-class machines have L1 and L2. L3 is rare, even on
big expensive Unix servers, but will become more common when CPUs start
coming with L1 and L2 on-chip.
Because most L1 caches are on the CPU chip, there's isn't very much room for
them so they tend to be small. It looks like the Pentium has 2 L1 caches, one
for instructions (I-cache) and one for data (D-cache); each is 8 KB. If this
is the only cache size available for Pentium, all laptops you look at will
have this.
The size of the L2 cache you get will depend on what brand and model of
laptop you buy, since Compaq and Fujitsu and NEC can decide independently how
much L2 cache to put on their motherboards (within a range defined by the CPU
chip). It's usually decided by the marketing people, not the technical
people, based on what chips are available at what prices and what price they
intend to sell the computer for. It looks like most benchmark results you'll
see are with 256 or 512 KB of L2 cache; AT&T makes one Pentium-based server
with 4 MB of L2 cache.
There are other cache-related buzzwords you may encounter.
"Write-back" means that when you update something in "memory" the cache
doesn't actually push the new value out to the memory chips (or to L2, if
it's an L1 write-back) until the "line" gets replaced in the cache. ("line"
is the chunk-size caches act on, usually a small number of bytes like 8, 16,
32, or 64 for L1, or 32, 64, 128, or 256 for L2)
"Write-through" means that when you update "memory" the cache updates its
value as well as sending an immediate update to physical memory (or to L2 if
it's an L1 write-through). Write-back is generally faster if your application
fits in the cache.
"Non-blocking, out of order" means that the CPU looks at the next N
instructions it's about to execute. It executes the first and finds that the
data isn't in cache. Since it's boring to just wait around for the data to
come back from memory, it looks at the next instruction. If that 2nd
instruction doesn't need the data the 1st instruction is waiting on, the CPU
goes ahead and executes that instruction. If the 3rd instruction does need
the data, it remembers it needs to execute that one after the data comes in
and goes on to the 4th instruction. Depending on how many outstanding
requests are allowed, if the 4th one causes a cache miss on a different line
it may put that one on hold as well and go on to the 5th instruction. The
Pentium Pro can do this, but I don't think the Pentium can.
"Set-associative" means the cache is split into 2 or more mini-caches.
Because of the way things are accessed in a cache, this can help a program
that has some "badly behaving" code mixed with some "good" code. Other terms
that go with it are "LRU" (the mini-cache picked for replacement is the one
Least Recently Used) or "random" (the line picked is selected randomly).
They can make a big difference in how happy you are with your system's
performance. There are enough variables that you probably aren't going to be
able to predict how happy you'll be with a configuration unless you sit down
in front of the machine and run whatever it is you plan to run on it. Make up
your own benchmark floppy with your primary application to take with you to
showrooms. (Throw it away after all your test drives, since it will probably
have collected a virus or three.)
Bigger or faster isn't always better. Speed is usually a tradeoff with size,
and you have to match L2 cache size/speed to CPU speed. A system with a
faster MHz CPU could perform worse than a system with a slower chip because
the CPU<-->L2 speed match might be such that the faster CPU requires a
different, slower mode on the L2 connection.
If all you want to do is run MyLittleSpreadsheet, and the code and data all
fit in 400 KB, a system with 512 KB of L2 cache will likely run more than
twice as fast as a system with 256 KB of L2. If MLS fits in 600 KB and has a
very sequential access pattern (a "cache-buster"), the 128 KB and 256 KB
systems will perform about the same -- like a dog; if the pattern is random
rather than sequential, the 512KB system will probably do some fractional
amount better than the 128 KB system. This is why it's so important to try
out your application and ignore impressive numbers for programs you're never
going to run.
Also, you may find the Doom benchmark page useful :-).
-----------------------------------------------------------------------------
4.3.2. How Caching Works
Caches get their leverage from exploiting two kinds of locality in memory
references. Temporal locality means "access now, it should be accessed again
soon", while spatial means "if byte N was asked for, byte N+1, N+2, N+3 will
probably be wanted too". Because Unix multitasks, every context switch
violates both types of locality. Spatial is impacted because contexts may not
be located closely together. Temporal is impacted because you have to wait
until all other ready contexts get their chance to run before you can run
again.
One side-effect of what's today considered "good programming practice", with
high-level languages using a lot of subroutine calls, is that the program
counter of a typical process hops around like crazy. You might think that
this, together with the periodic clock interrupts for multitasking, would
make spatial locality very poor.
However, the clock interrupt only fires about 60 times per second. This is a
very low overhead, if you consider how many instructions can be exectuted at
60 MHz in 1/60th of a second (for a poor estimate, something like 30 MIPS * 1
/60 = half a million instructions--at 16 bits each, roughly a megabyte of
memory has been walked through!). This is lots of opportunity to take
advantage of temporal locality -- and most programs are not so large that
their time-critical parts won't fit inside a megabyte.
(Thanks to Michael T. Pins and Joan Eslinger for much of this section.)
-----------------------------------------------------------------------------
4.3.3. A Short Primer on Cache Design
Before we go further in discussing specifics of the Intel processors we'll
need some basic cache-design theory. (Some of this repeats and extends the
Overview.)
Modern system designs have two levels of caching; a primary or internal cache
right on the chip, and a secondary or external cache in high-speed memory
(typically static RAM) off-chip. The internal cache feeds the processor
directly; the external cache feeds the internal cache.
A cache is said to hit when the processor, or a higher-level cache, calls for
a particular memory location and gets it. Otherwise, it misses and has to go
to main memory (or at least the next lower level of cache) to fetch the
contents of the location. A cache's hit rate is the percentage of time,
considered as a moving average, that it hits.
The external cache is added to reduce the cost of an internal cache miss. To
speed the whole process up, it must serve the internal cache faster than main
memory would be able to do (to hide the slowness of main memory). Thus, we
desire a very high hit rate in the secondary cache as well as very high
bandwidth to the processor.
Obviously, secondary cache hit rate can be improved by making it bigger. It
can also be increased by increasing the associativity factor (more on this
later, but for now note that too much associativity can cost a big penalty).
A cache is divided up into lines. Typically, in an i486 system, each line is
4 to 16 bytes long (the i486 internal cache uses 16-byte lines; external line
size varies). When the processor reads from an external-cache address that is
not in the internal cache, that address and the surrounding 16 bytes are read
into a line.
Each cache line has a tag associated with it. The tag stores the address in
memory that the data in that cache line came from. (Plus a bit to indicate
that this line contains valid data).
Some more important terms describing how caches interact with memory:
write-through
it wouldn't do to let your cache get out of sync with main memory. The
simplest way to handle this is to arrange that every write to cache
generates the corresponding write to main store. In the simplest kind of
"write-through" cache, then, you only get cache speedup on reads. But if
the cache system includes write postings, writes will usually be fast
too.
write posting
Most write-through cache designs have a few `write posting' buffers that
allow the cache to accept write data as fast as a write back cache for
later writing to memory. So, as far as the processor is concerned, most
writes will happen at zero wait states (the full cache speed), unless the
processor issues enough writes in a short interval to cause the write
posting buffers to fill up faster than they can be emptied.
write-back
For each cache address range in DRAM, writes are done to cache only until
a new block has to be fetched due to out-of-range access (at which point
the old one is flushed to DRAM). This is much faster, because you get
cache speedup on writes as well as reads. It's also more expensive and
trickier to get right.
Write-back secondary caches are generally not a good idea. Beyond what was
said in the write-through paragraph above, recall that the goal of the
secondary cache is to have a high hit rate and high bandwidth to the
processor's internal cache. When a cache-miss occurs in the secondary cache,
often the line being replaced is dirty and must be written to main memory
first. The total time to service the secondary cache miss nearly doubles.
Even when the secondary cache line being replaced is not dirty, the service
time goes up because the dirty bit must first be examined before accessing to
main memory. Write-through caches have the advantage of being able to look up
data in the secondary cache and in main memory in parallel (in the case where
the secondary cache misses, some of the delay of looking in main memory has
already been taken away). (Write-back caches cannot do this because they
might have to write-back the cache line before doing the main memory read.)
For these reasons, write-back caches are generally regarded as being inferior
to write-posting buffers. They cost too much silicon and more often than not
perform worse.
Now some terms that describe cache organization. To understand these, you
need to think of main RAM as being divided into consecutive, non-overlapping
segments we'll call "regions". A typical region size is 64K. Each region is
mapped to a cache line, 4 to 128 bytes in size (a typical size is 16). When
the processor reads from an address in a given region, and that address is
not already in core, the location and others near it are read into a line.
direct-mapped
Describes a cache system in which each region has exactly one
corresponding line (also called "one-way cache").
two-way set-associative
Each region has *two* possible slots; thus your odds of not having to
fetch from DRAM (and, hence, your effective speed) go up.
There are also "four-way" caches. In general, an n-way cache has n pages per
region and improves your effective speed by some factor proportional to n.
However, multiset caches become very costly in terms of silicon real estate,
so one does not commonly see five-way or higher caches.
Because set-associative caches make better use of SRAM, they typically
require less SRAM than a direct-mapped cache for equivalent performance.
They're also less vulnerable to Unix's heavy memory usage. Andy Glew of
USENET's comp.arch group says "the usual rule of thumb is that a 4-way
set-associative cache is equivalent to a direct-mapped cache of twice the
size". On the other hand, some claim that as cache size gets larger, two-way
associativity becomes less useful. According to this school of thought it
actually becomes a net loss over a direct-mapped cache at cahe sizes over
256K.
So, typically, you see multi-set cache designs on internal caches, but
direct-mapped designs for external caches.
The larger you make a cache line, the cheaper the design will be, as you save
on expensive tag ram, but the worse the performance will be, as you pay for
each cache miss manyfold. It is not reasonable to have 32 bytes reload on a
cache miss.
In the presence of interleaved DRAM memory, a cache line should not be larger
than a whole DRAM line -- double interleaved: 2*4 bytes, quadruple 4*4 bytes.
Otherwise, memory fetches to the external cache get slow.
An external cache which can support the i486 burst mode can increase
bandwidth to a much higher level than one which doesn't, and can
significantly reduce the cost of an internal cache miss.
-----------------------------------------------------------------------------
4.4. Suggestions for Buying
The best advice your humble editor can give is a collection of rules of
thumb. Your mileage may vary...
-----------------------------------------------------------------------------
4.4.1. Rule 1: Buy only motherboards that have been tested with Unix
One of DOS's many sins is that it licenses poor hardware design; it's too
brain-dead to stretch the cache system much. Thus, bad cache designs that
will run DOS can completely hose Unix, slowing the machine to a crawl or even
(in extreme cases) causing frequent random panics. Make sure your motherboard
or system has been tested with some Unix variant.
-----------------------------------------------------------------------------
4.4.2. Rule 2: Be sure you get enough cache.
If your motherboard offers multiple cache sizes, make sure you how much is
required to service the DRAM you plan to install.
Bela Lubkin writes: "Excess RAM [over what your cache can support] is a very
bad idea: most designs prevent memory outside the external cache's cachable
range from being cached by the 486 internal cache either. Code running from
this memory runs up to 11 times slower than code running out of fully cached
memory."
-----------------------------------------------------------------------------
4.4.3. Rule 3: "Enough cache" is at least 64K per 16MB of DRAM
Hardware caches are usually designed to achieve effective 0 wait state
status, rather than perform any significant buffering of data. As a general
rule, 64Kb cache handles up to 16Mb memory; more is redundant.
A more sophisticated way of determining cache size is to estimate the number
of processes you expect to be running simultaneously (ie 1 + expected load
average, let this value be N). Your external cache should be about N * 32k in
size. The justification for this is as follows: upon a context switch, it is
a good idea to be able to hold the entire i486 internal cache in the
secondary cache. For each process you would need something less than 8k * 4
(since it is 4-set associative, you need 32k to help map the conflicting
cache lines, and the extra cache left over (24k) should be plenty to help
improve the hit rate of the secondary cache when the internal cache misses.
The number of main memory accesses caused by context switching should be
reduced.
Of course, if you are going to be running programs with large memory
requirements (especially data), then a huge secondary cache would probably be
a big win. But most programs in the run queue will be small though (ls, cat,
more, etc.).
-----------------------------------------------------------------------------
4.4.4. Rule 4: If possible, max out the board's cache when you first buy
Bela continues: "Get the largest cache size your motherboard supports, even
if you're not fully populating it with RAM. The motherboard manufacturer buys
cache chips in quantity, knows how to install them correctly, and you won't
end up throwing out the small chips later when you upgrade your main RAM."
Gerard Lemieux qualifies this by observing that if adding SRAM increases the
external cache line size, rather than increasing the number of cache lines,
it's a lose. If this is the case, then an external cache miss could cost you
dearly; imagine how long the processor would have to wait if the line size
grew to 1024 bytes. If the cache has a poor hit rate (likely true, since the
number of lines has not changed), performance would deteriorate.
Also (he observes), spending an additional $250 for cache chips might buy you
2-3% in performance (even in Unix). You must ask yourself if it is really
worth it.
-----------------------------------------------------------------------------
4.4.5. Caveat
A lot of fast chips are held back by poor cache systems and slow memory. To
avoid trouble, cloners often insert wait states at the cache, slowing down
the chip to the effective speed of a much slower chip.
(Thanks to Guy Gerard Lemieux <lemieux@eecg.toronto.edu> for insightful
comments on an earlier version of this section.)
-----------------------------------------------------------------------------
4.5. Bus Wars
This is yet another area in which progress has simplified your choices a lot.
There used to be no fewer than four competing bus standards out there (ISA,
EISA, VESA/VLB, PCI, and PCMCIA). Now there are effectively just two -- PCI
for desktop/tower machines and PCMCIA for laptops.
-----------------------------------------------------------------------------
4.5.1. Bus Types
PCI is Intel's fast 64-bit bus for the Pentium. Many PCI boards are actually
PCI/ISA that supports both standards, so you can use less expensive ISA
peripherals and controllers. Beware, though; dual-bus boards lose about 10%
of their performance relative to single-bus PCI boards.
In the laptop market everything is PCMCIA. PCMCIA peripherals are about the
size of credit cards (85x54mm) and vary in thickness between 5 and 10mm. They
have the interesting feature that they can be hot-swapped (unplugged out and
plugged in) while the computer is on. However, they are seldom seen in
desktop machines. They require a special daemon to handle swapping; free
versions are now standard under Linux.
-----------------------------------------------------------------------------
4.5.2. Plug And Play
Many PCI cards have a feature called ``Plug and Play''. These cards negotiate
with the operating system at boot time for things like IRQs and DMA channels
-- they have no jumpers. Beware of these! Linux doesn't yet have full support
for Plug and Play, though there are support utilities available. (Of the
Microsoft OSs, only Windows 95 and up supports Plug and Play fully -- DOS
can't handle it at all and Windows 3.1 requires manual intervention).
-----------------------------------------------------------------------------
4.5.3. Historical Note
There used to be two ISA buses, the original 8-bit IBM PC and a 16-bit
compatible extension sometimes called "AT bus". The term ISA didn't come into
use until well into the lifetime of the latter. Here's a more complete list:
ISA
The original IBM PC bus architecture. The 8-bit version is completely
extinct. The 16-bit AT version is still alive but has been declared
obsolete by Intel in the PC99. specification.
MCA
Micro-Channel Architecture. A ``standard'' that IBM attempted to
promulgate, esp. in the PS/2 series of machines. While they tried to
claim it was faster/more efficient, it really was only marginally better
than ISA. Its real advantage--to IBM--was that it was a closed
architecture; they didn't publish the details of implementation as they
had on virtually everything for the XT/AT. It failed horribly; customers
didn't want to walk back into that trap.
EISA
Extended ISA. Required motherboard setup and manufacturer-provided
descriptions that got loaded into flash ROM. Manually. By the user.
Superseded by VLB years ago and now extinct.
VLB
VESA Local Bus. Usually seen on video cards providing high-speed graphics
data transfer, it was also being touted for other cards. VLB slots
accepted ISA and EISA cards and a further extension called the VESA
local-bus specification. Supplanted by PCI.
PCI
Peripheral Component Interconnect. The winner of the bus wars.
-----------------------------------------------------------------------------
4.6. Disk Wars: IDE vs. SCSI
4.6.1. Overview
Another basic decision is IDE vs. SCSI. Either kind of disk costs about the
same, but the premium for a SCSI card varies all over the lot, partly because
of price differences between VLB and PCI SCSI cards and especially because
many motherboard vendors bundle an IDE chipset right on the system board.
SCSI gives you better speed and throughput and loads the processor less, a
win for larger disks and an especially significant consideration in a
multi-user environment; also it's more expandable.
In terms of pure disk speed, IDE will always be faster, as they use the same
underlying disks, and IDE has less overhead. As fast as disks are getting
today, the difference is effectively noise. The real advantage of SCSI comes
from its extra brains. IDE uses polled I/O, which means that when you are
accessing the disk, the CPU isn't doing anything else. Most SCSI systems, on
the other hand, are DMA based, freeing up the system to do other things at
the same time. Hence, in terms of full system performance, SCSI is indeed
faster if you have good hardware and an intelligent OS.
Another important win for SCSI is that it handles multiple devices much more
efficiently. You can have at most two IDE devices; four for EIDE. SCSI
permits up to 7 (15 for Wide SCSI).
If you have two IDE (or ST506 or ESDI) drives, only one can transfer between
memory and disk at once. In fact, you have to program them at such a low
level that one drive might actually be blocked from seeking while you're
talking to the other drive. SCSI drives are mostly autonomous and can do
everything at once; and current SCSI drives are not quite fast enough to
flood more than half the SCSI bus bandwidth, so you can have at least two
drives on a single bus pumping full speed without using it up. In reality,
you don't keep drives running full speed all the time, so you should be able
to have 3-4 drives on a bus before you really start feeling bandwidth crunch.
Of course, IDE is cheaper. Many motherboards have IDE right on board now; if
not, you'll pay maybe $15 for an IDE adapter board, as opposed to $200+ for
the leading SCSI controller. Also, the cheap SCSI cabling most vendors ship
can be flaky. You have to use expensive high-class cables for consistently
good results. See Mark Sutton's horror story.
-----------------------------------------------------------------------------
4.6.2. Enhanced IDE
These days you seldom see plain IDE; souped-up variants are more usual. These
are "Enhanced IDE" (E-IDE) and "Fast AT Attachment" (usually ATA for short).
ATA is Seagate's subset of E-IDE, excluding some features designed to permit
chaining with CD-ROMs and tape drives using the new "ATAPI" interface (an
E-IDE extension; so far only the CD-ROMs exist); in practice, ATA and E-IDE
are identical.
You'll need to be careful about chaining in CD-ROMs and tape drives when
using IDE/ATA. The IDE bus sends all commands to all disks; they're supposed
to latch, and each drive then checks to see whether it is the intended
target. The problem is that badly-written drivers for CD-ROMs and tapes can
collide with the disk command set. It takes expertise to match these
peripherals.
Neither ATA nor E-IDE has the sustained throughput capacity of SCSI (they're
not designed to) but they are 60-90% faster than plain old IDE. E-IDE's new
``mode 3'' boosts the IDE transfer rate from IDE's 3.3MB/sec to 13.3MB/sec.
The new interface supports up to 4 drives of up to 8.4 gigabytes capacity.
E-IDE and ATA are advertised as being completely compatible with old IDE.
Theoretically, you can mix IDE, E-IDE and ATA drives and controllers any way
you like, and the worst result you'll get is conventional IDE performance if
the enhancements don't match up (the controller picks the lowest latch
speed). In practice, some IDE controllers (notably the BusLogic) choke on
enhanced IDE.
Accordingly, I recommend against trying to mix device types an an E-IDE/ATA
bus. Unfortunately, this removes much of E-IDE/ATA's usefulness!
E-IDE on drives above 540MB does automatic block mapping to fool the BIOS
about the drive geometry (avoiding limits in the BIOS type tables). They
don't require special Unix drivers.
Many motherboards now support ``dual EIDE'' channels, i.e. two separate [E]
IDE interfaces each of which can, theoretically, support two IDE disks or
ATA-style devices.
-----------------------------------------------------------------------------
4.6.3. SCSI Terminology
The following, by Ashok Singhal <ashoks@duckjibe.eng.sun.com> of Sun
Microsystems with additions by your humble editor, is a valiant attempt to
demystify SCSI terminology.
The terms ``SCSI'', ``SCSI-2'', and ``SCSI-3'' refer to three different
specifications. Each specification has a number of options. Many of these
options are independent of each other. I like to think of the main options
(there are others that I'll skip over because I don't know enough about them
to talk about them on the net) by classifying them into five categories:
-----------------------------------------------------------------------------
4.6.3.1. Logical: SCSI-1, SCSI-2, SCSI-3>
This refers to the commands that the controllers understand. Shortly after
SCSI first came out, the vendors agreed on a spec for a common comand set
called CCS. CCS was made a required part of the SCSI-2 standard. You should
be able to use a SCSI disk with a SCSI-2 card and vice-versa as long as they
both support CCS. Non-CCS SCSI devices aren't worth considering.
``SCSI-3'' is a superset of SCSI-2 including commands intended for CD-R and
streaming multimedia devices.
-----------------------------------------------------------------------------
4.6.3.2. Electrical Interface
��*�single-ended (max cable length 6 meters)
��*�differential (max cable length 25 meters)
This option is independent of command set, speed, and path width.
Differential is less common but allows better noise immunity and longer
cables. It's rare in SCSI-1 controllers.
For a PC you will probably always see single-ended SCSI controllers but if
you're shopping around for disks you might run across differential disks.
They will likely be more expensive than single-ended ones and will not work
on your single-ended bus.
-----------------------------------------------------------------------------
4.6.3.3. Handshake
��*�Asynchronous (acknowledge each word (8, 16 or 32 bits) transferred.
��*�Synchronous (multiple-word transfers permitted between ACKS).
Synchronous is faster. This mode is negotiated between controller and device;
modes may be mixed on the same bus. This is independent of command set, data
width, and electrical interface.
-----------------------------------------------------------------------------
4.6.3.4. Synchronous Speed (does not apply for asynchronous option)
Normal transfer speed is 5 megabytes/sec. The ``fast'' option (10 mb/sec) is
defined only in SCSI-2 and SCSI-3. Fast-20 (or ``Ultra'') is 20 mb/sec;
Fast-40 (or "Ultra-2") is 40MB/sec. The fast options basically defines
shorter timing parameters such as the assertion period and hold time.
The parameters of the synchronous transfer are negotiated between each target
and initiator so different speed transfers can occur over the same bus.
-----------------------------------------------------------------------------
4.6.3.5. Path width
The standard SCSI data path is 8 bits wide. The ``wide'' option exploits a
16- or 32-bit data path (uses 68-pin rather than 50-pin data cables). You
also get 4-bit rather than 3-bit device IDs, so you can have up to 16
devices. The wide option doubles or quadruples your transfer rate, so for
example a fast-20/wide SCSI link using 16 bits transfers 40mb/sec.
What are those ``LUN'' numbers you see when you boot up? Think of them as
sub-addresses on the SCSI bus. Most SCSI devices have only one ``logical''
device inside them, thus they're LUN zero. Some SCSI devices can, however,
present more than one separate logical unit to the bus master, with different
LUNs (0 through 7). The only context in which you'll normally use LUNs is
with CD-ROM juke boxes. Some have been marketed that offer up to 7 CD-ROMS
with one read head. These use the LUN to differentiate which disk to select.
(There's history behind this. Back in the days of EISA, drives were supposed
to be under the control of a separate SCSI controller, which could handle up
to 7 such devices (15 for wide SCSI). These drives were to be the Logical
Units; hence the LUN, or Logical Unit Number. Then, up to 7 of these SCSI
controllers would be run by the controller that we today consider the SCSI
controller. In practice, hardware cost dropped so rapidly, and capability
increased so rapidly, it became more logical to embed the controller on the
drive.)
-----------------------------------------------------------------------------
4.6.4. Avoiding Pitfalls
Here are a couple of rules and heuristics to follow:
Rule 1: Total SCSI cable length (both external and internal devices) must not
exceed six meters. For modern Ultra SCSI (with its higher speed) cut that to
three feet!
It's probably not a good idea to cable 20MB/s or faster SCSI devices
externally at all. If you must, one of our informants advises using a Granite
Digital ``perfect impedance'' teflon cable (or equivalent); these cables
basically provide a near-perfect electrical environment for a decent price,
and can be ordered in custom configurations if needed.
A common error is to forget the length of the ribbon cable used for internal
devices when adding external ones (that is, devices chained to the SCSI
board's external connector).
Rule 2: Both ends of the bus have to be electrically terminated.
On older devices this is done with removable resistor packs ?? typically
8-pin-inline widgets, yellow or blue, that are plugged into a plastic
connector somewhere near the edge of the PCB board on your device.
Peripherals commonly come with resistor packs plugged in; you must remove the
packs on all devices except the two end ones in the physical chain.
Newer devices advertised as having "internal termination" have a jumper or
switch on the PCB board that enables termination. These devices are
preferable, because the resistor packs are easy to lose or damage.
Rule 3: No more than seven devices per chain (fifteen for Wide SCSI).
There are eight SCSI IDs per controller. The controller reserves ID 7 or 15,
so your devices can use IDs 0 through 6 (or 0 through 14, wide). No two
devices can share an ID; if this happens by accident, neither will work.
The conventional ID assignments are: Primary hard disk = ID 0, Secondary hard
disk = ID 1, Tape = ID 2. Some Unixes (notably SCO) have these wired in. You
select a device's ID with jumpers on the PCB or a thumbwheel.
SCSI IDs are completely independent of physical device chain position.
Heuristic 1: Stick with controllers and devices that use the Centronics-style
50-pin connector. Internally these connectors are physically identical to
diskette cables. Externally they use a D50 shell. This "standard" connector
is common in the desktop/tower/rackmount-PC world, but you'll find lots of
funky DIN and mini-DIN plugs on devices designed for Macintosh boxes and some
laptops. Ask in advance and don't get burned.
Heuristic 2: For now, when buying a controller, go with an Adaptec xx42 or
one of its clones such as the BusLogic 542. (I like the BusLogic 946 and 956,
two particularly fast Adaptec clones well-supported under Linux.) The Adaptec
is the card everybody supports and the de-facto standard. Occasional
integration problems have been reported with Unix under Future Domain and
UltraStor cards, apparently due to command-set incompatibilities. At least,
before you buy these, make sure your OS explicitly supports them.
However: Beware the combination of an Adaptec 1542 with a PCI Mach32 video
card. Older (1.1) Linux kernels handled it OK, but all current ones choke.
Your editor had to replace his 1542 because of this, swearing sulphurously
the while.
Heuristic 3: You'll have fewer hassles if all your cables are made by the
same outfit. (This is due to impedence reflections from minor mismatches. You
can get situations where cable A will work with B, cable B will work with C,
but A and C aren't happy together. It's also non-commutative. The fact that
`computer to A to B' works doesn't mean that `computer to B to A' will work.
Heuristic 4. Beware Cheap SCSI Cables!
Mark Sutton tells the following instructive horror story in a note dated 5
Apr 1997:
I recently added an additional SCSI hard drive to my home machine. I bought
an OEM packaged Quantum Fireball 2 gig SCSI drive (meaning, I bought a drive
in shrinkwrap, without so much as mounting hardware or a manual. Thank God
for Quantum's web page or I would have had no idea how to disable termination
or set the SCSI ID on this sucker. Anyway, I digress...). I stuck the drive
in an external mounting kit that I found in a pile of discarded computer
parts at work and my that boss said I could have. (All 5 of my internal bays
were full of devices.)
Anyway, I had my drive, and my external SCSI mounting kit, I needed a cable.
I went into my friendly local CompUSA in search of a SCSI cable, and
side-by-side, on two hooks, were two "identical" SCSI cables. Both were 3
feet. Both had centronics to centronics connectors, both were made by the
same manufacturer. They had slightly different model numbers. One was $16.00,
one was $30.00. Of course, I bought the $16 cable.
Bad, I say, BAD BAD MISTAKE. I hooked this sucker up like so:
??????- ?????? ????????- ??????
|Internal|--|Adaptec|??|New Quantum|??|UMAX |
|Devices | |1542CF | ^ | Disk | ^ |Scanner|
??????- ????-- | ????????- | ??????
| |
New $16 cable Cable that came
with scanner.
Shortly after booting, I found that data all over my old internal hard drive
was being hosed. This was happening in DOS as well as in Linux. Any disk
access on either disk was hosing data on both disks, attempts to scan were
resulting in corrupted scans *and* hosing files on the hard disks. By the
time I finished swapping cables around, and checking terminations and
settings, I had to restore both Linux and DOS from backups.
I went back to CompUSA, exchanged the $16 cable for the $30 one, hooked it up
and had no more problems.
I carefully examined the cables and discovered that the $30 cable contained
24 individual twisted pairs. Each data line was twisted with a ground line.
The $16 cable was 24 data wires with one overall grounded shield. Yet, both
of these cables (from the same manufacturer) were being sold as SCSI cables!
You get what you pay for.
(Another correspondent guesses that the cheap cable probably said
``Macintosh'' on it. The Mac connector is missing most of its ground pins.)
-----------------------------------------------------------------------------
4.6.5. Trends to Watch For
Disks of less that 2GB capacity simply aren't being manufactured anymore;
there's no margin in them. Our spies tell us that all major disk makers
retooled their lines a while back to produce 540MB unit platters, which are
simply being stacked 2N per spindle to produce ranges of drives with roughly
1GB increments of capacity. The highest reasonably-priced drives are still
9GB (16 platters per drive), but you can get 23GB or even 45GB capacities
(these are probably packing 2.4GB per platter).
Average drive latency is inversely proportional to the disk's rotational
speed. For years, most disks spun at 3600 rpm; most high-performance disks
now spin at 7,200 rpm, and high-end disks like the Seagate Cheetah line are
moving to 10,000 rpm. These fast-spin disks run extremely hot; expect cooling
to become a critical constraint in drive design.
Drive densities have reached the point at which standard inductive read/write
heads are a bottleneck. In newer designs, expect to see magnetoresistive head
assemblies with separate read and write elements.
-----------------------------------------------------------------------------
4.6.6. More Resources
There's a USENET [http://www.cis.ohio-state.edu/hypertext/faq/usenet/scsi-faq
/top.html] SCSI FAQ. Also see the home page of the T10 committee that writes
SCSI standards.
There is a large searchable database of hard disk and controller information
at the PC DISK Hardware Database.
-----------------------------------------------------------------------------
4.7. Other Disk Decisions
Look at seek times and transfer rates for your disk; under Unix disk speed
and throughput are so important that a 1-millisecond difference in average
seek time can be noticeable.
-----------------------------------------------------------------------------
4.7.1. Disk Brands
An industry insider (a man who buys hard drives for systems integration) has
passed us some interesting tips about drive brands. He says the absolute
best-quality drives are the Hewlett-Packards but you will pay a hefty premium
for that quality.
The other top-tier manufacturers are Quantum and Seagate; these drives
combine cutting-edge technology with very aggressive pricing.
The second tier consists of Maxtor, Conner, and Western Digital.
Maxtor often leads in capacity and speed, but at some cost in other quality
measures. For example, many of the high-capacity Maxtor drives have serious
RFI emission problems which can cause high error rates. SCSI has built-in ECC
correction, so SCSI drives only take a performance hit from this; but it can
lead to actual errors from IDE drives.
Western Digital sells most of its output to Gateway at sweetheart prices; WD
drives are thus not widely available elsewhere.
The third tier consists of Fujitsu, Toshiba, and everyone else. My friend
observes that the Japanese, despite their reputation for process engineering
savvy, are notably poor at drive manufacturing; they've never spent the money
and engineering time needed to get really good at making the media.
If you see JTS drives on offer, run away. It is reliably reported that they
are horrible.
Just as a matter of interest, he also says that hard drives typically start
their life cycle at an OEM price around $400 each. When the price erodes to
around $180, the product gets turfed ?? there's no margin any more.
I've found a good cheap source for reconditioned SCSI disks at Uptime
Computer Support Services.
-----------------------------------------------------------------------------
4.7.2. To Cache Or Not To Cache?
Previous issues said "Disk cacheing is good, but there can be too much of a
good thing. Excessively large caches will slow the system because the
overhead for cache fills swamps the real accesses (this is especially a trap
for databases and other applications that do non-sequential I/O). More than
100K of cache is probably a bad idea for a general-purpose Unix box; watch
out for manufacturers who inflate cache size because memory is cheap and they
think customers will be impressed by big numbers." This may no longer be true
on current hardware; in particular, most controllers will interrupt a
cache-fill to fulfill a `real' read request.
In any case, having a large cached hard drive (particularly in the IDEs)
often does not translate to better performance. For example, Quantum makes a
210Mb IDE drive which comes with 256Kb cache. Conner and Maxtor also have
210Mb drives, but only with 64Kb caches. The transfer rate on the drives,
however, show that the Quantum comes in at 890Kb/sec, while the Maxtor and
Conner fly away at 1200Kb/sec. Clearly, the Conner and Maxtor make much
better use of their smaller caches.
However, it may be that any hardware disk cacheing is a lose for Unix! Scott
Bennett <bennett@mp.cs.niu.edu> reports a discussion on comp.unix.wizards:
"nobody found the hardware disk caches to be as effective in terms of
performance as the file system buffer cache...In many cases, disabling the
hardware cache improved system performance substantially. The interpretation
of these results was that the cacheing algorithm in the kernel was superior
to, or at least better tuned to Unix accesses than, the hardware cacheing
algorithms."
On the other hand, Stuart Lynne <sl@mimsey.com> writes:
Ok. What I did was to use the iozone program.
What this showed was that on my root disk in single user mode I could get
about 500kb for writing and 1000kb for reading a 10MB file. With the disk
cache disabled I was able to get the same for writing but only about 500kb
for reading. I.e. it appears the cache is a win for reading, at least if you
have nothing else happening.
Next I used a script which started up iozone in parallel on all four disks,
two to each of the big disks (three) and one on the smaller disk. A total of
seven iozone's competing with each other.
This showed several interesting results. First it was apparant that higher
numbered drives did get priority on the SCSI bus. They consistantly got
better throughput when competing against lower numbered drives. Specifically
drive 1 got better results than drive 0 on controller 0. Drive 4 got better
results than drive 3 on controller 1. All of the drives are high end Seagate
and have similiar characteristics.
In general with cache enabled the results where better for reading than
writing. When the cache was disabled the write speed in some cases went up a
bit and the read speed dropped. It would seem that the readahead in some
cases can compete with the writes and slow them down.
My conclusions are that we'll see better performance with the cache. First
the tendency is to do more reading than writing in your average Unix system
so we probably want to optimize that. Second if we assume an adequate system
cache slow writes shouldn't affect an individual process much. When we write
we are filling the cache and we don't usually care how long it takes to get
flushed. Of course we would notice it when writing very large files.
Thus (this is your humble editor again), I can only recommend experiment. Try
disabling the cache. Your throughput may go up!
-----------------------------------------------------------------------------
4.8. Tuning Your I/O Subsystem
(This section comes to us courtesy of Perry The Cynic, <perry@sutr.cynic.org
>. My own experience agrees pretty completely with his.)
Building a good I/O subsystem boils down to two major points: pick matched
components so you don't over-build any piece without benefit, and construct
the whole pipe such that it can feed what your OS/application combo needs.
It's important to recognize that ``balance'' is with respect to not only a
particular processor/memory subsystem, but also to a particular OS and
application mix. A Unix server machine running the whole TCP/IP server suite
has radically different I/O requirements than a video-editing workstation.
For the ``big boys'' a good consultant will sample the I/O mix (by reading
existing system performance logs or taking new measurements) and figure out
how big the I/O system needs to be to satisfy that app mix. This is not
something your typical Linux buyer will want to do; for one, the application
mix is not static and will change over time. So what you'll do instead is
design an I/O subsystem that is internally matched and provides maximum
potential I/O performance for the money you're willing to spend. Then you
look at the price points and compare them with those for the memory
subsystem. That's the most important trade-off inside the box.
So the job now is to design and buy an I/O subsystem that is well matched to
provide the best bang for your buck. The two major performance numbers for
disk I/O are latency and bandwidth. Latency is how long a program has to wait
to get a little piece of random data it asked for. Bandwidth is how much
contiguous data can be sent to/from the disk once you've done the first
piece. Latency is measured in milliseconds (ms); bandwidth in megabytes per
second (MB/s). Obviously, a third number of interest is how big all of your
disks are together (how much storage you've got), in Gigabytes (GB).
Within a rather big envelope, minimizing latency is the cat's meow. Every
millisecond you shave off effective latency will make your system feel
significantly faster. Bandwidth, on the other hand, only helps you if you
suck a big chunk of contiguous data off the disk, which happens rarely to
most programs. You have to keep bandwidth in mind to avoid mis-matching
pieces, because (obviously) the lowest usable bandwidth in a pipe constrains
everything else.
I'm going to ignore IDE. IDE is no good for multi-processing systems, period.
You may use an IDE CD-ROM if you don't care about its performance, but if you
care about your I/O performance, go SCSI.
Let's look at the disks first. Whenever you seriously look at a disk, get its
data sheet. Every reputable manufacturer has them on their website; just read
off the product code and follow the bouncing lights. Beware of numbers (`<
12ms fast!') you may see in ads; these folks often look for the lowest/
highest numbers on the data sheet and stick them into the ad copy. Not
dishonest (usually), but ignorant.
What you need to find out for a disk is:
1. What kind of SCSI interface does it have? Look for "fast", "ultra", and
"wide". Ignore disks that say "fiber" or "differential" (these are
specialty physical layers not appropriate for the insides of small
computers). Note that you'll often find the same disk with different
interfaces.
2. What is the "typical seek" time (ms)? Make sure you get "typical", not
"track-to-track" or "maximum" or some other measure (these don't relate
in obvious ways, due to things like head-settling time).
3. What is the rotational speed? This is typically 4500, 5400, 7200, or
10000 rpm (rotations per minute). Also look for "rotational latency" (in
ms). (In a pinch, average rotational latency is approx. 30000/rpm in
milliseconds.)
4. What is the `media transfer rate' or speed (in MB/s)? Many disks will
have a range of numbers (say, 7.2-10.8MB/s). Don't confuse this with the
"interface transfer rate" which is always a round number (10 or 20 or
40MB/s) and is the speed of the SCSI bus itself.
These numbers will let you do apple-with-apples comparisons of disks. Beware
that they will differ on different-size models of the same disk; typically,
bigger disks have slower seek times.
Now what does it all mean? Bandwidth first: the `media transfer rate' is how
much data you can, under ideal conditions, get off the disk per second. This
is a function mostly of rotation speed; the faster the disk rotates, the more
data passes under the heads per time unit. This constrains the sustained
bandwidth of this disk.
More interestingly, your effective latency is the sum of typical seek time
and rotational latency. So for a disk with 8.5ms seek time and 4ms rotational
latency, you can expect to spend about 12.5ms between the moment the disk
`wants' to read your data and the moment when it actually starts reading it.
This is the one number you are trying to make small. Thus, you're looking for
a disk with low seek times and high rotation (RPM) rates.
For comparison purposes, the first hard drive I ever bought was a 20MB drive
with 65ms seek time and about 3000RPM rotation. A floppy drive has about
100-200ms seek time. A CD-ROM drive can be anywhere between 120ms (fast) and
400ms (slow). The best IDE harddrives have about 10-12ms and 5400 rpm. The
best SCSI harddrive I know (the Seagate Cheetah) runs 7.8ms/10000rpm.
Fast, big drives are expensive. Really big drives are very expensive (that's
20GB+ drives as of this writing in August 1998). Really fast drives are
pretty expensive (that's about < 8ms right now). On the other end, really
slow, small drives are cheap but not cost effective, because it doesn't cost
any less to make the cases, ship the drives, and sell them.
In between is a `sweet spot' where moving in either direction (cheaper or
more expensive) will cost you more than you get out of it. The sweet spot
moves (towards better value) with time. Right now (August 1998), it's about
at 4GB drives, 8-10ms, 5400-7200rpm, fast or ultra SCSI. If you can make the
effort, go to your local computer superstore and write down a dozen or so
drives they sell `naked'. (If they don't sell at least a dozen hard drives
naked, find yourself a better store. Use the Web, Luke!) Plot cost against
size, seek and rotational speed, and it will usually become pretty obvious
which ones to get for your budget.
Do look for specials in stores; many superstores buy overstock from
manufacturers. If this is near the `sweet spot', it's often surprisingly
cheaper than comparable drives. Just make sure you understand the warranty
procedures.
Note that if you need a lot of capacity, you may be better off with two (or
more) drives than a single, bigger one. Not only can it be cheaper (2x4GB is
often cheaper than 1x9GB), but you end up with two separate head assemblies
that move independently, which can cut down on latency quite a bit (see
below).
Once you've decided which kind of drive(s) you want, you must decide how to
distribute them over one or more SCSI buses. Yes, you may want more than one
SCSI bus. (My current desktop machine has three.) Essentially, the trick is
to make sure that all the disks on one bus, talking at the same time, don't
exceed the capacity of that bus. At this time, I can't recommend anything but
an Ultra/Wide SCSI controller. This means that the attached SCSI bus can
transfer data at up to 40MB/s for an Ultra/Wide disk, 20MB/s for an Ultra/
narrow disk, and 10MB/s for a `fast SCSI' disk. These numbers allow you do do
your math: an 8MB/s disk will eat an entire bus on its own if it's `fast'
(10MB/s). Three 6MB/s ultra/narrow disks fit onto one bus (3x6=18MB/s<20MB/
s), but just barely. Two ultra/wide Cheetahs (12.8MB/s) will share an (ultra/
wide) bus (25.6<40), but they would collide on an ultra/narrow bus, and any
one Cheetah would be bandwidth constrained on a (non-ultra) `fast' bus (12.8
> 10).
If you find that you need two SCSI buses, you can go with `dual channel'
versions of many popular SCSI controller cards (including the Adaptec). These
are simply two controllers on one card (thus taking only one PCI slot). This
is cheaper and more compact than two cards; however, on some motherboards
with more than 3 PCI slots, using two cards may be somewhat faster (ask me
what a PCI bridge is :-).
How do you deal with slow SCSI devices - CD-ROMS, scanners, tape drives,
etc.? If you stick these onto a SCSI bus with fast disks, they will slow down
things a bit. You can either accept that (as in ``I hardly ever use my
scanner anyway''), or stick them onto a separate SCSI bus off a cheap
controller card. Or you can (try to) get an ATA version to stick onto that
inevitable IDE interface on your motherboard. The same logic applies to disks
you won't normally use, such as removables for data exchange.
If you find yourself at the high end of the bandwidth game, be aware that the
theoretical maximum of the PCI bus itself is 132MB/s. That means that a dual
ultra/wide SCSI controller (2x40MB/s) can fill more than half of the PCI
bus's bandwidth, and it is not advised to add another fast controller to that
mix. As it is, your device driver better be well written, or your entire
system will melt down (figuratively speaking).
Incidentally, all of the numbers I used are `optimal' bandwidth numbers. The
real scoop is usually somewhere between 50-70% of nominal, but things tend to
cancel out - the drives don't quite transfer as fast as they might, but the
SCSI bus has overhead too, as does the controller card.
Whether you have a single disk or multiple ones, on one or several SCSI
buses, you should give careful thought to their partition layout. Given a set
of disks and controllers, this is the most crucial performance decision
you'll make.
A partition is a contiguous group of sectors on the disk. Partitioning
typically starts at the outside and proceeds inwards. All partitions on one
disk share a single head assembly. That means that if you try to overlap I/O
on the first and last partition of a disk, the heads must move full stroke
back and forth over the disk, which can radically increase seek time delays.
A partition that is in the middle of a partition stack is likely to have best
seek performance, since at worst the heads only have to move half-way to get
there (and they're likely to be around the area anyway).
Whenever possible, split partitions that compete onto different disks. For
example, /usr and the swap should be on different disks if at all possible
(unless you have outrageous amounts of RAM).
Another wrinkle is that most modern disks use `zone sectoring'. The upshot is
that outside partitions will have higher bandwidth than inner ones (there is
more data under the heads per revolution). So if you need a work area for
data streaming (say, a CD-R pre-image to record), it should go on an outside
(early numbered) partition of a fast-rotating disk. Conversely, it's a good
convention to put rarely-used, performance-uncritical partitions on the
inside (last).
Another notes concerns SCSI mode pages. Each (modern) SCSI disk has a small
part of its disk (or a dedicated EEPROM) reserved for persistent
configuration information. These parameters are called `mode pages', for the
mechanism (in the SCSI protocol) for accessing them. Mode page parameters
determine, among others, how the disk will write-cache, what forms of error
recovery it uses, how its RAM cache is organized, etc. Very few configuration
utilities allow access to mode page parameters (I use FWB Toolkit on a Mac -
it's simply the best tool I know for that task), and the settings are usually
factory preset for, uh, Windows 95 environments with marginal hardware and
single-user operation. Particularly the cache organization and disconnect/
reconnect pages can make a tremendous difference in actual performance.
Unfortunately there's really no easy lunch here - if you set mode page
parameters wrong, you can screw up your data in ways you won't notice until
months later, so this is definitely `no playing with the pushebuttons'
territory.
Ah yes, caches. There are three major points where you could cache I/O
buffers: the OS, the SCSI controller, and the on-disk controller. Intelligent
OS caching is by far the biggest win, for many reasons. RAM caches on SCSI
controller cards are pretty pointless these days; you shouldn't pay extra for
them, and experiment with disabling them if you're into tinkering.
RAM caches on the drives themselves are a mixed bag. At moderate size
(1-2MB), they are a potential big win for Windows 95/98, because Windows has
stupid VM and I/O drivers. If you run a true multi-tasking OS like Linux,
having unified RAM caches on the disk is a significant loss, since the
overlapping I/O threads kick each other out of the cache, and the disk ends
up performing work for nothing.
Most high-performance disks can be reconfigured (using mode page parameters,
see above) to have `segmented' caches (sort of like a set-associative memory
cache). With that configured properly, the RAM caches can be a moderate win,
not because caching is so great on the disk (it's much better in the OS), but
because it allows the disk controller more flexibility to reschedule its I/O
request queue. You won't really notice it unless you routinely have >2 I/O
requests pending at the SCSI level. The conventional wisdom (try it both
ways) applies.
And finally I do have to make a disclaimer. Much of the stuff above is
shameless simplification. In reality, high-performance SCSI disks are very
complicated beasties. They run little mini-operating systems that are most
comfortable if they have 10-20 I/O requests pending at the same time. Under
those circumstances, the amortized global latencies are much reduced, though
any single request may experience longer latencies than if it were the only
one pending. The only really valid analysis are stochastic-process models,
which we really don't want to get into here. :-)
-----------------------------------------------------------------------------
4.9. Souping Up X Performance
If you care about X performance, be sure you get a graphics card with a
dedicated blitter and a high-speed local-bus connection. If it says "AGP" you
have this; AGP is a cross-vendor standard for a local bus optimized for
graphics.
These cards speed up X in two ways. First, they offload some common
screen-painting operations from the main processor onto specialized
processors on the card itself. Secondly, by using a local bus, they make it
possible to send commands to the card faster than the ISA bus could allow.
The combined effect can be eye-poppingly fast screen updates even at very
high resolutions.
There's no longer much reason to bother with any of the commercial X servers
like MetroLink or X/Inside. XFree86 now supports most of the high-end cards
that used to be the special preserve of the commercial X versions.
If you're feeling really flush, plump for a 15", 17" or even 20" monitor. The
larger size can make a major difference in viewing comfort. Also you'll be
set for 1600x1200, which many cards can support these days. In the mean time,
the bigger screen will allow you to use fonts in smaller pixel sizes so that
your text windows can be larger, giving you a substantial part of the benefit
you'd get from higher pixel resolutions.
-----------------------------------------------------------------------------
5. Hardware for Backups
You should have a tape drive for backup. Ideally, your tape backup should be
able to image your entire disk. Choosing a tape drive used to be pretty
complicated, with a plethora of different formats and media to chose from.
It's much simpler now that the combination of cheap CD-ROM drives and huge
hard disks has effectively killed off QIC and other sub-megabyte formats.
There are a bunch of non-tape niche technologies for backup, including
floptical disks, Bernoulli boxes, Iomega and SyQuest removable drives, and
magneto-optical drives. Ignore them all; they're half-assed attempts to
combine a backup device with the fast random access needed for working
storage that don't do either job very cost-effectively, especially when you
consider the (high) cost of their media. Only magneto-optical drives are
likely to have much of a future, and that only given improvements in access
speed.
Digital Data Storage (DDS) capacities are a good match for today's
multi-gigabyte drives (this is essentially the same technology as Digital
Audio Tape or DAT). I'm told that Hewlett-Packard DDS devices are especially
good, not surprising given HP's traditional obsession with reliability and
overengineering stuff. All the DDSs I know about are SCSI devices.
At the high end, 8mm helical-scan tape (the stuff used in Sony camcorders)
competes with DDS. This is a single-source tchnology, from Exabyte.
Capacities are 2.2 and 5 gig, transfer speeds up around 500Kbytes/sec.
However, a correspondent says ``Don't touch Exabyte. I've got three. All
three have been sent back for warranty repair at least once.'' He also says
``A significant expense can be the cleaning tapes. Exabyte is notorious for
this.'' So it's probably a good idea to stick with DDS if you have
high-capacity requirements.
OTOH, Carl Renneberg <renneber@sci-log.apana.org.au> says of the Exabyte that
his drive has proven to be rock steady and reliable, and recommends it with
the following provisos:
��*�You must, simply must, use the correct kind of tape. Do not use 8mm video
cartridges - they leave junk on the drive's head, and have dropouts. They
are simply not designed for storing data. Use only the data grade tape
from Sony (or from Exabyte - Sony manufactures the tape, Exabyte relabels
it).
��*�Use the correct tape cleaner. Like tape cartridges, it's cheaper to buy
the Sony brand than the Exabyte brand.
Carl has found that, where someone has had poor experience with Exabyte
drives, it's because the owner - or the previous owner - did not take care of
the drive, or used junk tapes.
Here's a quick summary of the major alternative DDS formats:
Table 2. DDS types
+--------------+----------------------+--------------------+----------------+
|Type |Megabytes |Megabytes |Speed (Kbytes/ |
| |(uncompressed) |(compressed) |sec) |
+--------------+----------------------+--------------------+----------------+
|DDS-1 60-meter|1300 |2K-4K |183-366 |
+--------------+----------------------+--------------------+----------------+
|DDS-1 90-meter|2000 |4K-8K |183-366 |
+--------------+----------------------+--------------------+----------------+
|DDS-2 |5000 |7K-12K |183-500 |
|120-meter | | | |
+--------------+----------------------+--------------------+----------------+
DDS tape drives (and tapes) come actually in four variants: DDS, DDS-DC,
DDS-2, and DDS-3. These are supposed to be downward compatible (e.g. DDS-2
reads/writes DDS-DC but not vice versa.) DDS and DDS-DC use 60m and 90m
tapes; the -DC version adds hardware compression. DDS (non-DC) should be
considered obsolete. DDS-2 adds 120m tapes and denser recording:
There is also a yet-newer DDS-3 standard, with yet again higher density on
the tape. DDS-3 is bleeding edge (high premium), but DDS-2 is coming down
now, and can make the difference between single-tape and and multi-tape
backups (which can often make the difference between daily backups and "why
didn't I..." hand-wringing.)
-----------------------------------------------------------------------------
6. Of Mice And Machines
Mice and trackballs used to be simple; now, thanks to Microsoft, they're
complicated. In the beginning, there was only the Mouse Systems 3-button
serial mouse; this reported status to a serial port 30 times a second using a
5-byte serial packet encoding now called ``C'' protocol. The Logitech Series
7 and 9 mice were Mouse Systems-compatible. All Unixes that have any mouse
support at all understand C-protocol serial mice.
Then Microsoft got into the act. They designed a two-button serial mouse
which reports only deltas in a three-byte packet; that is, it sends changes
in button status and motion reports only when the mouse is actually moving.
This is called `M' protocol. Microsoft sold a lot of mice, so Logitech
switched from `C' to `M' ?? but they added a third button, state changes for
which show up in an optional fourth byte. Thus, `M+' protocol,
upward-compatible with Microsoft's `M'. Most Unix vendors add support for M+
mice, but it's wise to check.
Bus mice are divided into 8255 and InPort types. These report info
continuously at 30 or 60 Hz (though InPort mice have an option for reporting
deltas only), and you get interrupts on events and then have to poll hardware
ports for details. Bus mice are no longer widely available and there is no
good reason to bother with them.
In addition to serial mice and bus mice, there are ``keyboard mice''. On PS/
2s there are two identical-looking keyboard ports, labeled (with icons)
``mouse'' & ``keyboard''. Both are 6-pin mini-DINs that look like the regular
PC keyboard port only smaller. Physically, the connector goes to the keyboard
processor (often an 8042); electrically, it provides TTL levels (serial
clock, serial data, ground and +5V); logically, it uses the same protocol as
normal serial mice. The same keyboard processor that decodes the keyboard
decodes the mouse. PS/2s have this port, many newer motherboards do as well.
All things considered, Unix users are probably best off going with a serial
mouse if their motherboard isn't one of the newer ones with a built-in PS/2
mouse. Most current clone motherboards give you two serial ports, so you can
still dedicate one to this and still have one for the all-important modem.
Not only are the compatibility issues less daunting, but a serial mouse loads
the multitasking system less due to interrupt frequency.
Beware that most clone vendors, being DOS oriented, bundle two-button M-type
mice for which Unix support is presently spotty, and they may not work with
your X. Thus, you may have to buy your own three-button mouse. Ignore the
adspeak about dpi and pick a mouse/trackball that feels good to your hand.
Your editor really, really likes the Logitech TrackMarble, an optical
trackball that eliminates the chronic roller-fouling problems of the older
TrackMan. They're well-supported by XFree86 (type MouseMan), so any Linux or
BSDI will accept them.
-----------------------------------------------------------------------------
7. Modems
This section will give you a thumbnail sketch of the modem types available
out there, one tuned for the typical Unix installation's needs.
-----------------------------------------------------------------------------
7.1. Overview of the Modem Market
The modem market is like consumer electronics (and unlike the computer market
as a whole) in that price is a very poor predictor of performance. For
ordinary file transfers, some $50 modems are better than some $150 modems.
Paying top dollar mainly buys you better tolerance of poor connections and
better performance at heavy-duty bi-directional transfers (such as you would
generate, for exmaple, using SLIP or PPP over a leased line to an Internet
provider).
In today's market, the typical modem does a nominal 56kbps -- V.90 plus V.29
or V.17 fax transmission and reception. You don't see much in the way of slow
/cheap to fast/expensive product ranges within a single brand, because
competition is fierce and for many modem board designs (those featuring DSP
(Digital Signal Processor) chips run by a program in ROM) adding a new
protocol is basically a software change.
Detailed discussion of the V-series standards can be found in the [glossary]
Glossary. At this point all the jargon is mainly of historic interest.
For much more information on high-speed modems for Unix, see The Linux/Modem
Compatibility Knowledge Base. The page is aimed at Linux users, but the
advice is general to any non-Windows OS.
-----------------------------------------------------------------------------
7.2. Internal vs. External
Most modems come in two packagings: internal, designed to fit in a PC card
slot, and external, with its own case, power supply, and front-panel lights.
Typically you'll pay $20 to $30 more for an external modem than you will for
the internal equivalent. You'll also need a serial port to connect your
external modem to.
Pay that premium ?? being able to see the blinkenlights on the external ones
will help you understand and recover from pathological situations. For
example, if your Unix system is prone to ``screaming-tty'' syndrome, you'll
quickly learn to recognize the pattern of flickers that goes with it. Punch
the hangup/reset button on an external modem and you're done ?? whereas with
an internal modem, you have to go root and flounder around killing processes
and maybe cold-boot the machine just to reset the card.
See Rick's Rants for extended discussion of this point.
-----------------------------------------------------------------------------
7.3. Pitfalls to Avoid
Beware of hype. Some modems with 14400 or higher FAX transmission rates have
a lower speed-limit on data transmissions. If you're in any doubt, get a
quote of the ``data rate'', not FAX bit rate.
The best way to avoid problems is to make the vendor tell you what V-series
standards the product supports. Then use the handy [glossary.html] glossary
above.
If the abbreviation ``RPI'' occurs anywhere on the box, don't even consider
buying the modem. RPI (Rockwell Protocol Interface) is a proprietary
``standard'' that allows modem makers to save a few bucks at your expense by
using a cheap-jack Rockwell chipset that doesn't do error correction.
Instead, it hands the job off to a modem driver which (on a Unix machine) you
will not have.
Also avoid anything called a ``Windows Modem'' or ``WinModem'', ``HCF, or
``HSP''; these lobotomized pieces of crap require a Windows DLL to run. They
will eat up to 25% of your processor clocks during transfers, and hog
high-priority interrupts (causing your machine to stall under Windows even if
your processor still has spare cycles).
All modems advertise mainly on DCE speed. Get a quote for max DTE speed too,
if you can; make sure the DTE rate is enough times faster to handle your
maximum on-the-fly compression (with V.42bis that's 4:1).
Multi-user Unix eats enough processor clocks that you want to be sure of good
hardware buffering in your UART ?? that is, enough of it to avoid losing
characters between modem and PC if the OS is a bit slow responding to an
interrupt (V.42bis in hardware won't detect this!). This means you want a
16550A or equivalent UART. If you're using an external modem, this is an
issue about your serial-port board(s). If you're using an internal modem, the
UART is on the modem card itself. So, when buying internal modems, ask what
the UART type is. If the vendor says 16540, lose them.
Many fax modems come with bundled MS-DOS fax software that is at best useless
under Unix, and at worst a software kluge to cover inadequate hardware. Avoid
these bundles and buy a bare modem ?? it's cheaper, and lowers the likelihood
that something vital to your communications needs has been left out of the
hardware.
Avoid ``Class 1'' and ``Class 2'' modems. Look for ``Class 2.0'' for the full
EIA-standard command set.
-----------------------------------------------------------------------------
7.4. A Modem Glossary
An explanation of modem jargon. None of this is very interesting any more in
the age of ubiquitous cheap V.90 modems.
First, some ancient history:
Bell 103
AT&T 110bps standard. Dead as a doornail. Anyone tries to sell you a Bell
103 modem, report 'em to the bunco squad. The damn things were the size
of breadboxes and flakier than bad dandruff. Not totally useless, though;
they make decent space heaters.
Bell 202
AT&T standard for 300bps transmission. Every modem in existence does this
one (or, at least, if you find one that doesn't, donate it to an
archeological museum instantly).
Bell 212A
AT??T standard for 1200bps transmission. Yes, a few people are still
chugging gamely away with 1200bps modems.
MNP 2, MNP 3, MNP 4
Microcom Network Protocol, a series of error-detection-and-correction
standards primarily used for 1200 and 2400-baud modems. Now incorporated
in V.42.
MNP 5
Microcom Network Protocol Level 5, a an error-correction and
data-compression method that runs on top of MNP 2-4, now displaced by
V.42bis. (Most modems that support V.42bis also support the MNP 5 and
auto-detect it in use by other modems).
Now a new term enters the picture:
CCITT
International Consultative Committee on Telephony and Telegraphy (the
acronym is from the French official name of the organization, Comit�
Consultatif Internationale T�l�graphique et T�l�phonique). This is the
standards body for modems, fax equipment and related gear. Now known as
ITU-T (International Telecommunications Union- Telecommunication
Standards Sector).
ITU-T is responsible for the "V" series of modem specifications, These are:
V.22
ITU-T standard for 1200bps. Not quite compatible with Bell 212A, thus
never used much in the U.S. This caused a few trans-border snafus in the
early 1980s. Most modems still offer a V.22 compatibility mode, but
nobody doing international data transfers settles for 1200bps any more,
so it's a dead issue.
V.22bis
ITU-T standard for 2400bps. AT&T got out of the modem-standards business
after 212A; thus, V.22 was the first open standard to be supported
worldwide.
V.25ter
ITU-T standard for the AT command set.
V.32
The ITU-T standard for 9600bps transmission.
V.32bis
The ITU-T standard for 12200 and 14400bps data transmission.
V.42
The ITU-T standard for firmware error correction between V.32 and V.32bis
modems (also worked with V22 and V22bis). Helps them cope with noisy
lines.
V.42bis
The ITU-T standard for V.42 with on-the-fly compression. Same bit speeds
as V.32bis and V.42, but a higher effective transfer rate (especially for
text, graphic images and other high-redundancy data, for which it may
manage up to 4:1 compression).
V.34
ITU-T standard for 28800bps transmission. Also known as V.fast before the
standard was ratified in June 1994. Amended in 1996 to support 33600bps
transmission.
V.90
ITU-T recommendation for asymmetric data signalling rates of up to 56Kbps
in the direction of a digitally connected server to a capable client, and
up to 33.6Kbps in the direction of the client to the server.
The technology is based on eliminating restrictions imposed by the
conversion of analog signals to digital form in the downstream data path
(server -> client). Data flow in the server to client direction does not
occur in the form of a modulated carrier, it is instead sent as binary
numbers representative of 256 possible voltage levels. The reason for the
asymmetrical send/receive rates is because in the direction from the
client to the server it is not possible to use a digital coding scheme
and make it work as well as V.34 does, thus V.34 is used instead. It
isn't possible because the telco's line card has a codec that is a much
better digital level changer for the transmit direction than it is for
the receive direction. The codec used in the customer's modem is, in that
respect, somewhat more sophisticated and was designed to work as a fairly
good level changer in the receive direction (which the telco's codec was
not designed to do).
Note: Achievable bit rates are limited to less than 56kbps in the United
States by FCC regulations that limit power input to the network.
These ITU-T protocols form a neat speed and capability ladder. Usually, any
modem that can do V.32bis can do V.32; any modem that can do V.42 can do
V.32bis; any modem that can do V.42bis can do V.42; any modem that can do
V.34 can do V.32; and so forth. In fact, modems in this family automatically
recognize each other and negotiate for the highest speed both can handle.
Beyond the V series, the telcos are pushing a technology called ADSL
(Asymmetric Digital Subscriber Line). This method converts existing
twisted-pair telephone lines into access paths for multimedia and high speed
data communications. ADSL transmits more than 6Mbps to a subscriber, and as
much as 640 kbps more in both directions.
An ADSL circuit connects an ADSL modem on each end of a twisted-pair phone
line, creating three information channels; a high speed downstream channel, a
medium speed duplex channel, and a POTS (Plain Old Telephone Service)
channel. The POTS channel is split off from the digital modem by filters,
thus guaranteeing uninterrupted POTS, even if ADSL fails. The high speed
channel ranges from 1.5 to 6.1 Mbps, while duplex rates range from 16 to 640
kbps. Each channel can be sub-multiplexed to form multiple, lower rate
channels.
Now to the vendor-proprietary encodings:
V.32terbo
A non-ITU-T ``standard'' for 16800 and 19200bps transmission floated by a
vendor consortium frustrated with the glacial pace of the V.fast
development. V.terbo was intended as a stopgap until V.fast comes in;
technically it's very close to V.32bis. Obsolete since V.34 was ratified.
HST
Proprietary 16.8kcps protocol used by US. Robotics modems. Never very
important in the Unix world, and now being eclipsed by the V-series
modems.
PEP
Packet Exchange Protocol, used for 19200bps between Telebit modems. These
modems were very widely used in the Unix/USENET/UUCP community at one
time (for years, Telebit offered substantial discounts and/or two-for-one
deals to buyers with Internet or UUCP addresses). They were
super-reliable, programmable, fast as all get out, famously good at
pushing bits through noisy and lossy phone lines, and died stone dead
because they were a single-source product designed around a proprietary
protocol (and correspondingly expensive).
TurboPEP
Enhanced PEP, 115200bps ?? faster than V.fast even dreams of (but don't
expect to get quite that full speed unless your connection is really
clean).
56Kbps Modems
[Pre-V.90] - Rockwell, USR, Lucent Technologies, and Motorola marketed
incompatible chipsets/modems that operated in a server/client format at
up to 56Kbps over standard telephone lines prior to the adoption of ITU-T
V.90. USR implemented a protocol dubbed X2, and the remainder combined
efforts to implement a protocol dubbed K56Flex (a combination of
Rockwell's K56Plus and Lucent's VFlex/2 protocols). The X2 and K56Flex
protocols do not interoperate.
Finally, some important modem features:
DCE
Data Communications Equipment rate, the maximum modem-to-modem bit speed.
DTE
Data Terminal Equipment rate, maximum computer-to-modem bit speed. If
your modem does N-to-1 compression, be sure your DTE rate is at least N
times your DCE rate, or you won't be able to keep the modem fed fast
enough to get top DCE speed.
And one important misfeature:
RPI
Rockwell Protocol Interface -- a scam designed to save manufacturer a few
bucks by omitting error-correction logic from V-series modems. Instead
it's done in software by a proprietary driver. Do not buy such a modem,
it won't fly under Unix. See the Pitfalls to Avoid section.
Fax capability is included with most all modems these days; it's cheap for
manufacturers, being basically a pure software add-on. The CCITT also sets
fax protocol standards. Terms to know:
V.29
CCITT standard for Group III fax encoding at 9600bps
V.17
CCITT standard for Group III fax encoding at 14400bps
V.17 isn't common yet, but it doesn't usually cost extra over V.29 when you
find it.
There's a separate series of standards for software control of fax modems
over the serial line maintained by the Electronics Industry Association and
friends. These are:
Class 1 ?? base EIA standard for fax control as extensions to the Hayes AT
command set.
Class 2.0 ?? enhanced EIA standard including compression, error correction,
station ID and other features.
Class 2 ?? marketroidian for anything between Class 1 and Class 2.0.
Different ``Class 2'' modems implement different draft subsets of the 2.0
standard, so ``Class 2'' fax software won't necessarily drive any given
``Class 2'' modem.
There's also a proprietary Intel "standard" called CAS, Communicating
Applications Specification. Ignore it; only Intel products support it.
The GNU toolset includes a freeware fax transmission and reception toolset,
Netfax. Look for it at prep.ai.mit.edu:pub/gnu/fax-*. It says it requires a
modem conforming to the ``Class 2'' control standard, but you'd be safest
getting a 2.0-conformant modem for reasons explained above. Netfax also
requires GNU Ghostscript to do Postscript handling for it.
-----------------------------------------------------------------------------
8. CD-ROMs and Multimedia Hardware
8.1. CD-ROM Drives
Standard CD-ROMs hold about 650 megabytes of read-only data in a format
called ISO-9660 (formerly ``High Sierra''). All current Unixes now support
these devices. In fact, most Unix and Linux software is now distributed on
ISO-9660 CD-ROM, a cheaper and better method than the QIC tapes we used to
use.
CD-ROM drives may be driven through SCSI, through enhanced IDE (ATAPI), or
through a proprietary interface card (like the Mitsumi and Sony interfaces).
Unix support for CD-ROMs is usually through SCSI drivers (exception: BSDI/386
supports the Mitsumi interface; Linux supports the Mitsumi and Sony
interfaces). I recommend that you avoid the proprietary cards; they will
effectively cost you money when you need to upgrade. Besides, street prices
for SCSI CD-ROMS have dropped below $200.
(A few external CD-ROMs come with a parallel-port interface. Avoid these;
they tend to have very slow transfer rates.)
Any CD-ROM you buy should be at least a ``double-spin'' drive meeting the
MPC2 (Multimedia PC) standard of a 300K/sec transfer rate when reading.
digital data. The older single-speed drives, which only supported the 150K/
sec rate Red Book standard for audio CDs, are obsolete. The lowest speed you
can buy these days is 4X (600K/sec). 6X, 8X, 10X, 12X, 24X, and 32X are
available.
The next level up in CD hardware standards is CD-ROM XA. So far, drives that
support XA are few and expensive. It's not yet in wide use in the DOS/Windows
world, and I don't know of any Unix support for it, either in commercial or
freeware code.
CD-ROM access times about 280ms for high-end double-speed drives (to put this
in perspective, it's about 30 times slower than a typical 9ms hard disk, but
considerably faster than a tape). Accordingly, modern 32X drives are about
half the speed of a hard drive.
Many CD-ROM drives use a caddy (a plastic frame into which you drop the CD
before inserting both in the drive slot). Some are caddyless. The caddy has
two advantages: (1) it helps protect the CD from scratches, and (2) it
pre-positions the and supports the CD for the drive spindle, allowing faster
controlled rotation. Consequently, most of the faster CD-ROM drives use
caddies. They have the disadvantage of making CD-changing slightly more
awkward.
Most CD-ROMS will include a headphone jack so you can play audio CDs on them.
Better-quality ones will also include two RCA jacks for use with speakers.
Another feature to look for is a drive door or seal that protects the drive
head from dust.
CD-ROM formats are still an area of some confusion. A slight enhancement of
the original ``High Sierra'' CD-ROM filesystem format (designed for use with
DOS, and limited to DOS's 8+3 file-naming convention) has been standardized
as ISO-9660.
There is a de-facto Unix standard called `Rock Ridge' pioneered by the Sun
User's Group shareware CD-ROMs. This is a way of putting an extra layer of
indirection on an ISO-9660 layout that preserves Unix's long dual-case
filenames. Some Unixes (notably Linux, netBSD, freeBSD and BSD/OS) can mount
Rock Ridge filesystems.
More much more detail on CD-ROMs, CD-ROM standards and how to buy drives is
available in the alt.cdrom FAQ, available for FTP as [ftp://cdrom.com/cdrom/
faq] cdrom.com:/cdrom/faq. It is also archived in the news.answers tree at
rtfm. This FAQ includes comparison tables tables of numerous drive types,
CD-ROM sources, and ordering information.
-----------------------------------------------------------------------------
8.2. Sound Cards and Speakers
Look for the following features as a minimum in your sound card:
��*�16-bit sampling (for 16,5536 dynamic levels rather than 256).
��*�Mono and stereo support.
��*�Sampling rates ranging fron 8K/sec (voice-quality) through 11KHz
(AM-radio quality), 22KHz (FM-radio quality) and standard audio
(44.1KHz).
��*�MIDI interface via a standard 15-pin D-shell connector.
��*�RCA output jacks for headphones or speakers.
��*�A microphone jack for sound input.
The most important feature is your sound card is what type of MIDI synthesis
it has. All current sound cards have sort of MIDI (musical instrument digital
interface) compliance, but if nothing else, make sure your card supports the
General MIDI standard. MIDI is just a set of commands issued by the
application that tells the sound card which instrument to play, at what note,
and for what duration.
Older and cheaper cards use FM synthesis. This synthesis uses a combination
of sine waves to imitate the sounds of the different instruments. The result
is like the sound tracks of most computer games sold a few years ago;
imitation music with an arcade-like sound.
The method used by most modern sound boards is called wave table synthesis.
In this method, digitized samples of actual instrument sounds are used as
templates for the tones generated by the MIDI commands. Wave table cards vary
in the quantity and quality of samples; one figure of merit often quoted is
the wave table ROM size (often 4MB or 8MB). Also some boards have wavetable
RAM that can store samples loaded from a disk.
Soundcards with DSP (Digital Signal Processing) can perform synthesis effects
on board, relieving the CPU for other tasks. Some DSP chips are even
software-programmable. Some high-end cards even include 3D sound effects.
Whether the system used is SRS (Sound Retrieval System), Q-Sound, or
Spatializer, it is designed to improve the perceived stereo effect of your
speakers. These 3D effects work by delaying the timing of certain portions of
the audio signal so that different frequencies hit your ear at slightly
different times. The downside is that some of the cards equipped with 3D
sound add a noticeable amount of noise to the card's output.
If you play a lot of computer games, you'll need to pay attention to
compatibility. DOS games are written almost exclusively for the Creative Labs
specification; you will need a card that is 100% Sound Blaster compatible.
Many vendors do not license the Creative Labs specification but claim that
their cards are 100% game compatible. This means that the sound will work,
but not all sounds that you hear will be the ones that the game programmers
intended. If you play many DOS games, it would be best to buy a Sound Blaster
and save yourself a migraine.
Lastly, try to avoid sound cards with built-in amplifiers that are more
powerful than 4 watts/channel. Sound cards that have more powerful amplifiers
are said to have the problem of adding noise to the card's output. Use
powered speakers with a 4 watt/channel card to solve this problem. Most cards
are equipped with 4 watts/channel anyway. Wavetable cards are so inexpensive
these days that it's almost worth their additional cost over a regular FM
synthesis card. If you decide to settle for an FM card, make sure that there
is a daughterboard made for the card that will let you upgrade to wavetable
synthesis. In some cases, however, the wavetable card is cheaper than buying
an FM card and then deciding that you want the wavetable upgrade. If you do
decide on the wavetable as your card of choice, PC Magazine rated the best
MIDI wavetables (MIDI being the most important feature in my opinion) the
Media Vision Premium 3-D, Media Vision Pro 3-D, Creative Labs Sound Blaster
AWE32, and the Turtle Beach Monterey (although there are value editions of
the Sound Blaster 32 that have fewer ROM instrument samples but maintain the
superior MIDI wavetable synthesis).
Linux, includes drivers for the SoundBlaster series and many other boards
including the PAS Adlib, the Gravis Ultrasound, and the ProAudioSpectrum.
Many of these are included in the 2.0 and newer kernels. See the OSS/Free
home page for details
Warning: some sound cards require a specific CD-ROM type! Avoid these.
In speakers, look for a magnetically-shielded enclosure with volume, bass and
treble controls. Some speakers run off the card's 4-watt signal; others are
``self-powered'', using batteries or a separate power supply. Your major
buying choice is which one of these options to pursue.
If you're interested in music composition and playback (as opposed to just
sound effects) another important buying choice is FM versus wave-table
synthesis. FM is cheaper, wave-table is better. In-depth discussion of this
tradeoff is beyond the scope of this document. For more details, see the
PCsoundboards/generic FAQ (available on rtfm in the news.answers archive).
Generally speaking, you'll get better value if you buy your sound card,
CD-ROM, and speakers not separately but as a bundled ``multimedia upgrade
kit'' (even though Unix users will usually have to throw out the bundled
software).
One final, important tip: that audio cable from your CD-ROM back to the sound
card is used only when you play audio CD-ROMs through your speakers.
Software-generated sound goes through the system bus, so you can play
``Doom'' with sound even if your sound board won't accept the audio cable
connector.
-----------------------------------------------------------------------------
9. Special considerations when buying laptops
Up until about 1999 the laptop market was completely crazy. The technology
was in a state of violent flux, with ``standards'' phasing in and out and
prices dropping like rocks. Things are beginning to settle out a bit more
now.
One sign of this change is that there are now a couple of laptop lines that
are clear best-of-breeds for reasons having as much to do with good
industrial design and ergonomics as the technical details of the processor
and display.
In lightweight machines, I'm a big fan of the Sony VAIO line. I've owned one
from early 1999 until it physically disintegrated under the rigors of travel
in late 2000, and could hardly imagine switching. They weigh 3.5 pounds, give
you an honest 3 hours of life per detachable battery pack, have a very nice
1024x768 display, and are just plain pretty. Their only serious drawback is
that they're not rugged, and often fall apart after a year or so of use.
If you want a full-power laptop that can compete with or replace your desktop
machine, the IBM ThinkPad line is the bomb. Capable, rugged, and nicely
designed (though somewhat heavyweight for my taste).
These machines are not cheap, though. If you're trying to save money by
buying a no-name laptop, here are things to look for:
First: despite what you may believe, the most important aspect of any laptop
is not the CPU, or the disk, or the memory, or the screen, or the battery
capacity. It's the keyboard feel, since unlike in a PC, you cannot throw the
keyboard away and replace it with another one unless you replace the whole
computer. Never buy any laptop that you have not typed on for a couple hours.
Trying a keyboard for a few minutes is not enough. Keyboards have very subtle
properties that can still affect whether they mess up your wrists.
A standard desktop keyboard has keycaps 19mm across with 7.55mm between them.
If you plot frequency of typing errors against keycap size, it turns out
there's a sharp knee in the curve at 17.8 millimeters. Beware of ``kneetop''
and ``palmtop'' machines, which squeeze the keycaps a lot tighter and
typically don't have enough oomph for Unix anyway; you're best off with the
"notebook" class machines that have full-sized keys.
Second: be careful that your laptop meets the minimum core and disk
requirements for the Unix you want to run. This is generally not a problem
with desktop machines, which can be upgraded cheaply and easily, but laptops
often have more stringent constraints.
Third: with present flatscreens, 1024x768 color is the best you're going to
do (though that may change soon). If you want more than that (for X, for
example) you have to either fall back to a desktop or make sure there's an
external-monitor port on the laptop (and many laptops won't support higher
resolution than the flatscreen's).
Fourth: look for Nickel-Metal-Hydride (NiMH) batteries, as opposed to the
older (Nickel-Cadmium) NiCad type. NiMH batteries are great because they have
considerably higher energy capacity per pound that NiCads. They need special
circuitry to charge them fast, so don't try to throw out your NiCads and
replace them with NiMH cells if you use a fast charger intended for NiCads.
Both kinds of cells can be damaged by overcharging at rates faster than 10
hours per full charge.
Fifth: Older laptop electronics are still 5-volt CMOS. Most current designs
are 3.3-volt CMOS with power-management features on the processor (these are
often labelled APM, Advanced Power Management). Buy this, if you can, to
nearly double your use time between recharges.
Sixth: about those vendor-supplied time-between-recharge figures; don't
believe them. They collect those from a totally quiescent machine, sometimes
with the screen or hard disk turned off. Under DOS, you'd be lucky to get
half the endurance they quote; under Unix, which hits the disk more often, it
may be less yet. Figures from magazine reviews are more reliable.
Seventh: You probably want a color dual-scan display. It used to be that you
had to choose between passive-matrix LCD (cheap, miserable color) and active-
matrix LCD (great color, horribly expensive). Dual-scan passive-matrix is
nearly as good as active-matrix, except for the narrower viewing angle, and
it's much cheaper. Avoid the older single-scan models, sometimes marketed as
having STN (super-twisted nematic) displays.
Eighth: get either a CD-ROM drive or an Ethernet card. Otherwise initial load
of your Unix could turn into a serious problem...
-----------------------------------------------------------------------------
10. How to Buy
10.1. When to Buy
It used to be that good configurations for Unix were what the market called
`server' machines, with beefed-up I/O subsystems and fast buses. No longer;
today's `servers' are monster boxes with multiple power supplies and
processors, gigabytes of memory, and industrial-grade air cooling -- they're
not really suitable as personal machines. A typical SCSI desktop workstation
is as much as you'll need.
Prices keep dropping, so there's a temptation to wait forever to buy. A
tactic that makes a lot of sense in this market, if you have the leisure, is
to fix in your mind a configuration and a trigger price that's just a little
sweeter than the market now offers and buy when that's reached.
Before you shop, do your homework. Publications like "Computer Shopper" (and
their web site at [http://www.computershopper.com] http://
www.computershopper.com) are invaluable for helping you get a feel for prices
and what clonemakers are doing. Another excellent site is [http://
www.computeresp.com] ComputerESP.
-----------------------------------------------------------------------------
10.2. Where to Buy
The most important where-to-buy advice is negative. Do not go to a
traditional, business-oriented storefront dealership. Their overheads are
high. So are their prices.
Especially, run -- do not walk -- away from any outfit that trumpets
`business solutions'. This is marketing code for the kind of place that will
justify a heavy price premium by promising after-sale service and training
which, nine times out of ten, will turn out to be nonexistent or incompetent.
Sure, they'll give you plush carpeting and a firm handshake from a guy with
too many teeth and an expensive watch -- but did you really want to pay for
that?
There are two major alternatives to storefront dealerships and one minor one.
The major ones are mail order and computer superstores. The minor one is
computer fairs.
-----------------------------------------------------------------------------
10.3. Computer Fairs
I used to be a big fan of hole-in-the-wall stores run by immigrants from the
other side of the International Date Line, but most of those places have been
driven out of the regular retail game by the superstores. The only place you
find diaspora Chinese and Indians selling cheap PCs over the counter anymore
is at computer fairs. (Usually they're doing it to publicize a mail-order
business.)
You can find good `loss-leader' deals on individual parts at these fairs
(they're especially good places to buy disk drives cheap). But I call them a
minor alternative because it's hard to get a custom SCSI-based configuration
tuned for Unix built for you at a fair. So you end up, effectively, back in
the mail-order or Web channel.
-----------------------------------------------------------------------------
10.4. Mail Order
Direct-mail or Web buying makes a lot of sense today for anyone with more
technical savvy than J. Random Luser in a suit. Even from no-name mail-order
houses, parts and system quality tend to be high and consistent, so
conventional dealerships don't really have much more to offer than a warm
fuzzy feeling. Furthermore, competition has become so intense that even
mail-order vendors today have to offer not just lower prices than ever before
but warranty and support policies of a depth that would have seemed
incredible a few years back. For example, many bundle a year of on-site
hardware support with their medium- and high-end "business" configurations
for a very low premium over the bare hardware.
Note, however, that assembling a system yourself out of mail-order parts is
not likely to save you money over dealing with the mail-order systems houses.
You can't buy parts at the volume they do; the discounts they command are
bigger than the premiums reflected in their prices. The lack of any system
warranty or support can also be a problem even if you're expert enough to do
the integration yourself ?? because you also assume all the risk of defective
parts and integration problems.
Cruise through `Computer Shopper' and similar monthly ad compendia. Even if
you decide to go with a conventional dealer, this will tell you what their
premiums look like.
Watch out for dealers (Spectrum Trading for one) who charge ridiculous
shipping fees. One of our spies reports he bought a hotswappable hard disc
drive tray that weighed about 3 lbs. and cost $250 and they charged $25 to
ship it UPS groud.
Don't forget that (most places) you can avoid sales tax by buying from an
out-of-state mail-order outfit, and save yourself 6-8% depending on where you
live. If you live near a state line, buying from a local outfit you can often
win, quite legally, by having the stuff shipped to a friend or relative just
over it. Best of all is a buddy with a state-registered dealer number; these
aren't very hard to get and confer not just exemption from sales tax but
(often) whopping discounts from the vendors. Hand him a dollar afterwards to
make it legal.
(Note: I have been advised that you shouldn't try the latter tactic in
Florida -- they are notoriously tough on "resale license" holders).
(Note II: The Supreme Court has ruled that states may not tax out-of-state
businesses under existing law, but left the way open for Congress to pass
enabling legislation. Let's hope the mail-order industry has good lobbyists.)
If you can afford to trade a little extra money for minimum hassles, you
should go with a Linux systems integrator like [http://www.valinux.com] VA
Linux Systems. These guys will sell you tested and tuned hardware with Linux
pre-loaded. What you get for your price premium includes a support line to
call.
-----------------------------------------------------------------------------
10.5. Computer Superstores
Big chain superstores like CompUSA and Circuit City give you a reasonable
alternative to mail order. And there are good reasons to explore it -- these
stores buy and sell at volumes that allow them to offer prices not far above
mail-order.
One thing you should not buy mail-order if you can avoid it is a monitor.
Monitors are subject to substantial quality variations even within the same
make and model. Also, one good bump during shipping can twist the yoke on a
monitor so the image is tilted with respect to the bezel. So buy your monitor
face-to-face, picking the best out of three or four.
Another good argument for buying at a superstore is that you may have to pay
return postage if you ship a mail-order system back. On a big, heavy system,
this can eat your initial price savings.
The only major problem with superstores is that the salespeople who staff
them aren't very bright or very clueful (it's a sort of Darwinian
reverse-selection effect; these are the guys who are fascinated by computer
technology but not smart enough to be techies). Most of them don't know from
Linux and are likely to push things like Plug'n'Play cards and two-button
mice and (worse!) controllerless modems, that you can't use. Use caution and
check your system manifest.
-----------------------------------------------------------------------------
10.6. Other Buying Tips
You can often get out of paying tax just by paying cash, especially at
computer shows. You can always say you're going to ship the equipment out of
the state.
A lot of vendors bundle DOS or Windows and variable amounts of apps with
their hardware. If you tell them to lose all this useless cruft they may
shave $50 or $100 off the system price.
-----------------------------------------------------------------------------
11. Questions You Should Always Ask Your Vendor
11.1. Minimum Warranty Provisions
The weakest guarantee you should settle for in the mail-order market should
include:
��*�72-hour burn-in to avoid that sudden infant death syndrome. (Also, try to
find out if they do a power-cycling test and how many repeats they do;
this stresses the hardware much more than steady burn-in.)
��*�30 day money-back guarantee. Watch out for fine print that weakens this
with a restocking fee or limits it with exclusions.
��*�1 year parts and labor guarantee (some vendors give 2 years).
��*�1 year of 800 number tech support (many vendors give lifetime support).
Additionally, many vendors offer a year of on-site service free. You should
find out who they contract the service to. Also be sure the free service
coverage area includes your site; some unscrupulous vendors weasel their way
out with "some locations pay extra", which translates roughly to "through the
nose if you're further away than our parking lot".
If you're buying store-front, find out what they'll guarantee beyond the
above. If the answer is "nothing", go somewhere else.
-----------------------------------------------------------------------------
11.2. Documentation
Ask your potential suppliers what kind and volume of documentation they
supply with your hardware. You should get, at minimum, operations manuals for
the motherboard and each card or peripheral; also an IRQ list. Skimpiness in
this area is a valuable clue that they may be using no-name parts from Upper
Baluchistan, which is not necessarily a red flag in itself but should prompt
you to ask more questions.
-----------------------------------------------------------------------------
11.3. A System Quality Checklist
There are various cost-cutting tactics a vendor can use which bring down the
system's overall quality. Here are some good questions to ask:
��*�Is the memory zero-wait-state? One or more wait states allows the vendor
to use slower and cheaper memory but will slow down your actual memory
subsystem throughput. This is a particularly important question for the
cache memory!
��*�If you're buying a factory-configured system, does it have FCC
certification? While it's not necessarily the case that a non-certified
system is going to spew a lot of radio-frequency interference,
certification is legally required ?? and becoming more important as clock
frequencies climb. Lack of that sticker may indicate a fly-by-night
vendor, or at least one in danger of being raided and shut down! (For
further discussion, see the section on Radio Frequency Interference
above.)
��*�Are the internal cable connectors keyed, so they can't be put in upside
down? This doesn't matter if you'll never, ever ever need to upgrade or
service your system. Otherwise, it's pretty important; and, vendors who
fluff this detail may be quietly cutting other corners.
-----------------------------------------------------------------------------
12. Things to Check when Buying Mail-Order
12.1. Tricks and Traps in Mail-Order Warranties
Reading mail-order warranties is an art in itself. A few tips:
Beware the deadly modifier ``manufacturer's'' on a warranty; this means you
have to go back to the equipment's original manufacturer in case of problems
and can't get satisfaction from the mail-order house. Also, manufacturer's
warranties run from the date they ship; by the time the mail-order house
assembles and ships your system, it may have run out!
Watch for the equally deadly ``We do not guarantee compatibility''. This
gotcha on a component vendor's ad means you may not be able to return, say, a
video card that fails to work with your motherboard.
Another dangerous phrase is ``We reserve the right to substitute equivalent
items''. This means that instead of getting the high-quality name-brand parts
advertised in the configuration you just ordered, you may get those no-name
parts from Upper Baluchistan ?? theoretically equivalent according to the
spec sheets, but perhaps more likely to die the day after the warranty
expires. Substitution can be interpreted as ``bait and switch'', so most
vendors are scared of getting called on this. Very few will hold their
position if you press the matter.
Another red flag: ``Only warranted in supported environments''. This may mean
they won't honor a warranty on a non-DOS system at all, or it may mean
they'll insist on installing the Unix on disk themselves.
One absolute show-stopper is the phrase ``All sales are final''. This means
you have no options if a part doesn't work. Avoid any company with this
policy.
-----------------------------------------------------------------------------
12.2. Special Questions to Ask Mail-Order Vendors Before Buying
��*�Does the vendor have the part or system presently in stock? Mail order
companies tend to run with very lean inventories; if they don't have your
item in stock, delivery may take longer. Possibly much longer.
��*�Does the vendor pay for shipping? What's the delivery wait?
��*�If you need to return your system, is there a restocking fee? and will
the vendor cover the return freight? Knowing the restocking fee can be
particularly important, as they make keep you from getting real
satisfaction on a bad major part. Avoid dealing with anyone who quotes
more than a 15% restocking fee ?? and it's a good idea, if possible, to
avoid any dealer who charges a restocking fee at all.
Warranties are tricky. There are companies whose warranties are invalidated
by opening the case. Some of those companies sell upgradeable systems, but
only authorized service centers can do upgrades without invalidating the
warranty. Sometimes a system is purchased with the warranty already
invalidated. There are vendors who buy minimal systems and upgrade them with
cheap RAM and/or disk drives. If the vendor is not an authorized service
center, the manufacturer's warranty is invalidated. The only recourse in case
of a problem is the vendor's warranty. So beware!
-----------------------------------------------------------------------------
12.3. Payment Method
It's a good idea to pay with AmEx or Visa or MasterCard; that way you can
stop payment if you get a lemon, and may benefit from a buyer-protection plan
using the credit card company's clout (not all cards offer buyer-protection
plans, and some that do have restrictions which may be applicable). However,
watch for phrases like ``Credit card surcharges apply'' or ``All prices
reflect 3% cash discount'' which mean you're going to get socked extra if you
pay by card.
Note that many credit-card companies have clauses in their standard contracts
forbidding such surcharges. You can (and should) report such practices to
your credit-card issuer. If you already paid the surcharge, they will usually
see to it that it is returned to you. Credit-card companies will often stop
dealing with businesses that repeat such behavior.
-----------------------------------------------------------------------------
12.4. Which Clone Vendors to Talk To
12.4.1. Some picks
If you've got the bucks to pay for high-end, high-quality hardware, I don't
know of anybody better than [http://www.varesearch.com] VA Linux Systems.
(Full disclosure: They've given me hardware and made me a director of the
company -- but all long after I got real happy with the way they do things.)
-----------------------------------------------------------------------------
12.4.2. Some pans
Dell: treated the Unix community, customers and its own employees very badly
back in 1994 by making an internal decision to kill its market-leading SVr4
port, then obfuscating and lying about its intentions for months after its
actions made the direction clear. On the other hand, they started getting
serious about Linux in early 2000. I won't say boycott them the way I used
to, but I will say you ought to think about VA instead.
Gateway: may also be a vendor to avoid. Apparently their newer machines don't
have parity bits in their memories; memory is tested only on reboot. This is
dubious design even for DOS, and totally unacceptable for Unix.
-----------------------------------------------------------------------------
13. Software to go with your hardware
I used to maintain an entire separate FAQ on Unixes for 386/486 and Pentium
hardware. Times change, industries evolve, and I can now replace that FAQ
with just three words:
Go get Linux!
-----------------------------------------------------------------------------
14. Other Resources on Building Linux PCs
Andrew Comech's [http://www.math.sunysb.edu/~comech/tools/CheapBox.html] The
Cheap /Linux/ Box page is a useful guide to building with current hardware
that is updated every two weeks. Andrew also maintains a [http://
www.math.sunysb.edu/~comech/tools/CheapBox.html#sample] short-cut version.
The Caveat Emptor guide has an especially good section on evaluating monitor
specifications.
Dick Perron has a PC Hardware Links page. There is lots and lots of good
technical stuff linked to here. Power On Self Test codes, manufacturer
address lists, common fixes, hard disk interface primer, etc.
Anthony Olszewski's Assembling A PC is an excellent guide to the perplexed.
Not Linux-specific. If you're specifically changing a motherboard, see the
[http://www.atipa.com/InfoSheets/instmb.shtml] Installing a Motherboard page.
This one even has a Linux note.
Tom's Hardware Guide covers many hardware issues exhaustively. It is
especially good about CPU chips and motherboards. Full of ads and
slow-loading graphics, though.
The System Optimization Site has many links to other worthwhile sites for
hardware buyers.
Christopher B. Browne has a page on [http://www.hex.net/~cbbrowne/
linuxvars.html] Linux VARs that build systems. He also recommends the Linux
VAR HOWTO.
Jeff Moe has a Build Your Own PC page. It's more oriented towards building
from parts than this one. Less technical depth in most areas, but better
coverage of some including RAM, soundcards and motherboard installation.
Features nifty and helpful graphics, one of the better graphics-intensive
pages I've seen. However, the hardware-selection advice is out of date.
The Linux Hardware Database provides, among other things (e.g., drivers,
specs, links, etc.), user ratings for specific hardware components for use
under Linux. Our ratings take a lot of the guess work out of choosing which
hardware to buy for a Linux box. The site also provides several
product-specific resources (i.e., drivers, workarounds, how-to) that help
users get hardware working after they have made a purchase.
-----------------------------------------------------------------------------
15. Upgrading Older Machines
15.1. Older Memory Types
The last-generation memory package was a SIMM (Single In-Line Memory Module).
We include some information on these here for people upgrading old systems.
SIMMs have 9 bits wide (8 bits of data, one for parity) and either 1, 4, or
16 megabytes in size. The 16s are slightly cheaper per megabyte.
There are two physical SIMM sizes; 30-pin and 72-pin. The 30-pin size is long
obsolete; most SIMM-using motherboards have exclusively 72-pin sockets.
Also, your memory may or may not have on-board parity check bits. Memory with
parity is specified as ``x36'' (4 check bits per word); memory without is
``x32''. Some motherboards require parity memory; some (including those based
on the Neptune chipset) can use parity if present; some (including those
based on the Triton chipset) cannot use it. Parity memory is a good idea for
Unix machines, which stress the hardware more than other OSs.
You'll hear a lot about ``EDO'' (Extended Data Out) RAM. This stuff is 5%-10%
more expensive, but that pays for a significant performance boost in ``burst
mode'' (that is, when fetching large contiguous sections of memory into the
cache). Most current motherboards (including those based on the Triton
chipset) can use EDO but don't require it.
You do not want to buy 1MB SIMMs any more, unless you're filling out memory
in an existing system configured with 1MB SIMMs. The trouble with these
(besides the fact that four of them are more expensive than a 4MB chip) is
that, due to interleaving constraints, you'll probably have to throw them
away if you ever add more memory.
These interleaving constraints differ from motherboard to motherboard.
Typically, memory sockets are set up in banks of either 2 or 4, with a
requirement that you fill all sockets of each bank with the same size of
chip. Also, if you mix chip sizes in the banks, the sizes have to decrease
smoothly (so there won't be gaps in the memory address space).
So your possible configurations look like the following table. All of these
are possible on 4-bank, 16-slot boards; only the asterisked ones are possible
on 2-bank, 8-slot boards. All the capacities above 64MB require that the
board accept 16MB SIMMs. All current 2x8 boards accept these.
* 4MB = 1-1-1-1
* 8MB = 1-1-1-1 + 1-1-1-1
12MB = 1-1-1-1 + 1-1-1-1 + 1-1-1-1
16MB = 1-1-1-1 + 1-1-1-1 + 1-1-1-1 + 1-1-1-1
* 16MB = 4-4-4-4
* 20MB = 4-4-4-4 + 1-1-1-1
24MB = 4-4-4-4 + 1-1-1-1 + 1-1-1-1
28MB = 4-4-4-4 + 1-1-1-1 + 1-1-1-1 + 1-1-1-1
* 32MB = 4-4-4-4 + 4-4-4-4
36MB = 4-4-4-4 + 4-4-4-4 + 1-1-1-1
40MB = 4-4-4-4 + 4-4-4-4 + 1-1-1-1 + 1-1-1-1
48MB = 4-4-4-4 + 4-4-4-4 + 4-4-4-4
52MB = 4-4-4-4 + 4-4-4-4 + 4-4-4-4 + 1-1-1-1
64MB = 4-4-4-4 + 4-4-4-4 + 4-4-4-4 + 4-4-4-4
* 64MB = 16-16-16-16
* 68MB = 16-16-16-16 + 1-1-1-1
72MB = 16-16-16-16 + 1-1-1-1 + 1-1-1-1
76MB = 16-16-16-16 + 1-1-1-1 + 1-1-1-1 + 1-1-1-1
* 80MB = 16-16-16-16 + 4-4-4-4
84MB = 16-16-16-16 + 4-4-4-4 + 1-1-1-1
88MB = 16-16-16-16 + 4-4-4-4 + 1-1-1-1 + 1-1-1-1
100MB = 16-16-16-16 + 4-4-4-4 + 4-4-4-4 + 1-1-1-1
112MB = 16-16-16-16 + 4-4-4-4 + 4-4-4-4 + 4-4-4-4
* 128MB = 16-16-16-16 + 16-16-16-16
132MB = 16-16-16-16 + 16-16-16-16 + 1-1-1-1
136MB = 16-16-16-16 + 16-16-16-16 + 1-1-1-1 + 1-1-1-1
144MB = 16-16-16-16 + 16-16-16-16 + 4-4-4-4
148MB = 16-16-16-16 + 16-16-16-16 + 4-4-4-4 + 1-1-1-1
160MB = 16-16-16-16 + 16-16-16-16 + 4-4-4-4 + 4-4-4-4
192MB = 16-16-16-16 + 16-16-16-16 + 16-16-16-16
196MB = 16-16-16-16 + 16-16-16-16 + 16-16-16-16 + 1-1-1-1
208MB = 16-16-16-16 + 16-16-16-16 + 16-16-16-16 + 4-4-4-4
256MB = 16-16-16-16 + 16-16-16-16 + 16-16-16-16 + 16-16-16-16
As you can see from the table, most configurations with 1MB chips in them
can't be upgraded without throwing away chips.
Here's some more about memory packaging.
DIP
Dual Inline Pack. These are the little 16-, 18- or more legged chips that
looked vaguely like squared-off centipedes. Populating 8Mb was a task for
an evening. Obsolete, except for cache memory.
SIPP
Single Inline Pin Pack. These look like SIMMs, but have round in-line
pins instead of the flat card-edge. Obsolete.
SIMM
Single Inline Memory Module. A small 30-pin card composed of several
chips (the number differs by manufacturer, age, and density) and support
components (mostly older modules) that installs in a snap-in socket. They
evolved through 128K, 256K, and 512K modules--all sadly obsolete
today--into 1Mb, 2Mb (rather rare), and 4Mb modules. Now generally
superseded by...
72-pin SIMM
Introduced in the last couple of years to cut down on the number modules
needed, they usually run 4Mb, 8Mb, 16Mb, and 32Mb. Many motherboards are
available with both 30 and 72-pin SIMM sockets to allow migration use of
your older SIMMs.
DIMM
Dual In-Line Memory Module. This is the current cutting edge, available
in 64MB and 128MB sizes.
There are also adapter cards that will accept 30-pin SIMM modules and plug
into a 72-pin socket; make sure you've the room for these if you intend to
try them, as they protrude quite a bit above the board. There are also
companies that will, for modest fees, desolder the memory chips from 30-pin
SIMMs and wave solder them onto 72-pin boards.
(Thanks to Dave Ihnat for the glossary.)
-----------------------------------------------------------------------------
15.2. If You Must Buy a 486
There were a bunch of early 386 boards that tanked Unix badly, due to timing
problems or botched cache design (one of these mistakes actually sank a major
vendor, Everex, when it got a piece of a big federal-government Unix contract
and then couldn't perform because its motherboards wouldn't run Unix
reliably!). Incompatibilities that gross are pretty much history now, unless
some low-ball vendor slips you a fossil from his back shelf. A good, tricky
way to avoid this is to specify a motherboard that can take 4 or 16MB SIMMs
(as opposed to just the older 1MB kind). You want to do this anyhow for
functional reasons.
Beware the infamous FP exception bug! Some motherboards fail to handle
floating point exceptions correctly; instead of generating a SIGFPE they lock
up. The following fragment of C code will reproduce the problem:
double d;
d = 0.0;
d = 1.0 / d; /* floating divide by zero should yield SIGFPE */
John R. Levine <johnl@iecc.cambridge.ma.us> explains: "The difficulty stems
from the fact that there are two ways to handle floating exceptions on a 486,
the right way and the PC way. What the 486 wants to do is to generate an
interrupt 16 when there is a floating point error, all entirely internal to
the CPU. This has been the native way to handle floating point interrupts
since the 286/287. The 286/287 and 386/387 each have a dedicated ERROR pin
that the FPU uses to tell the CPU that it's time for an error interrupt.
Unfortunately, the 8086/8087 handled interrupts differently. The error pin on
the 8087 was wired to the 8259A interrupt controller, the same interrupt
controller that handled keyboard, disk, clock, etc. interrupts. The PC/AT
enshrined IRQ 13 as the one for floating interrupts. (The details of this are
a little hazy to me, since the XT didn't have IRQ 13 tied to an 8259A, so the
AT must have at least changed the interrupt number.) PC designs have
generally wired the 287 or 387 ERROR pin to the 8259A, not to the ERROR pin
on the CPU, or at best had some poorly documented way to switch between the
two interrupt methods.
In the interest of backward compatibility, the 486 has a mode bit that says
not to handle FP exceptions automatically, but rather to freeze the FPU and
send a signal on the FERR pin, which is usually tied to an 8259A which then
feeds the interrupt back as IRQ 13. There is some extra complication involved
here because the FPU has to stay frozen until the interrupt is accepted so
the CPU can go back and look at the FPU's state. Early 386/25 chips had a bug
that would sometimes freeze up on a floating point interrupt and you had to
get a kludge socket with a PAL that fixed the timing glitch that provoked the
bug.
So as likely as not, the motherboard hardware that runs FERR out and back
isn't working correctly. It's not surprising, few DOS users take floating
point seriously enough to notice whether the interrupts are working right."
When you specify a system, make clear to your vendor that the motherboard
must handle float exceptions properly. Test your motherboard's handling of
divide-by-zero; if it doesn't work, press your vendor to replace it and send
me email! Only by publishing a list of boards known to be bad can we protect
ourselves and pressure vendors to fix this problem.
Norbert Juffa <s_juffa@iravcl.ira.uka.de> adds: Actually, the IBM PC, PC/XT
and most compatible use the NMI (non-maskable interrupt) to report
coprocessor errors. They don't go through the interrupt controller. Only a
few not quite compatible machines did use the 8259 PIC and one needed special
startup code for Microsoft-C for example to ensure correct handling of
coprocessor interrupts in programs. The PC/AT and compatibles do use the
8259, and the coprocessor interrupt comes in as INT 75h (IRQs from second
[slave] 8259 are mapped to INT 70h-77h) to the CPU. On the PC/XT it comes in
as INT 2 (NMI). The problem with using the NMI was that NMI is also used for
other purposes (e.g. parity error reporting) and that the service routine has
to figure out what really caused the interrupt. The reason not to use the
8259 on the PC might have been that not enough IRQs were available. The AT
has two cascaded 8259 chips and therefore has more IRQs available.
Bill Reynolds <bill@goshawk.lanl.gov> recommends that, if you're buying an
EISA motherboard, you check with the vendor to make sure it does not use the
`Hint' chipset. It isn't true EISA. A note in the back of the Hint manual
admits that the Hint's DMA, interrupts and timers aren't ISA-compatible
(however, Linux will run on it). Caveat emptor.
-----------------------------------------------------------------------------
15.3. Cache Engineering on 486 Machines
The 8K primary cache (L1) on the 486 is write-through, four-way
set-associative with a four-deep write posting buffer. It's been argued that
the 486 is more limited by that shallow write-posting buffer than by cache
miss overhead.
The 486's 8K internal primary cache is typically supplemented with an
external caching system (L2) using SRAM to reduce the cost of an internal
cache miss; in November 1994, 20ns SRAM is typical.
What varies between motherboards is the design of this secondary cache. 486
External Caches are Usually Direct-Mapped
Given the realities of the clone market, where buyers have been trained to
assume that a bigger number is always better, it is also easier to simply
advertise the 128K of direct mapped cache.
Because the 486 has write-posting buffers, a write-through external cache is
OK. Remember that the i486 already has a 4-deep write-buffer. Writes are done
when there is little bus activity (most common case) or when the buffer is
full. The write-buffer will write out its contents while the i486 CPU is
still crunching away from the internal cache. Thus, a write-through has
little negative impact on write performance. If the buffer was usually full,
then the i486 would be stalling while the write is done (and you would then
want a write-back cache OR a better choice would be another external write
buffer).
Also, recall that one of the goals of the secondary cache is to have very
high effective bandwidth to the i486 processor. Any associativity greater
than direct-mapped measurably increases the lookup time in the secondary
cache and increases the time to service an internal cache miss, and thus
reduces the effective bandwidth to the i486. Thus, direct-mapped secondary
caches provide a lower $$ cost AND increase the performance in the common
case of a secondary cache hit, even though the hit rate has been slightly
reduced by not adding the associativity.