1.1. Purpose of this document

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 Internet/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 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. In 2001 I published an article on building the Ultimate Linux Box.

This Buyer's Guide 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.

It may be a matter of historical interest that the page count of this guide peaked in mid-2001 and has been declining since. Video, sound, and other functions are migrating onto motherboards. Several bus types have disappeared, as have all the old-school backup technologies that couldn't scale up to match disk capacities, Spec sheets are getting simpler. Accordingly, there are parts that used to have whole sections to hemselves that I barely even write about anymore — mice, floppy disks, CD-ROM drives, and keyboards, for example, are utterly generic now,

Another consequence of the technology stabilizing is also that I'm updating this guide less often than I used to. Years can now go by without the PC market changing in any fundamental way.

In retrospect, the success of the ATX standard for motherboards in 1998-1999 was probably the turning point. The PC industry has become sufficiently commoditized that your choices are now getting simpler rather than more complicated. This is a Good Thing.

1.2. New versions of this document

New versions of the Unix Hardware Buyer HOWTO will be periodically 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.tldp.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 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 Hardware-HOWTO. It lists hardware known to be compatible with Linux, and hardware known to be incompatible. I've also done a series of articles on The Ultimate Linux Box.

3.1. Things to Not Care About

An effect of PC commoditization is that there aren lots of things you used to have to worry about that don't matter any more, because the market has completely flattened out. We list these here to get them out of the way.

---

3.1.2. Memory

Judging the memory-controller and cache design used to be one of the trickiest parts of evaluating a motherboard, but that stuff is all baked into the processor itself now. This removed a large source of latency and design variations. It also killed off the plethora of different RAM types that used to be out there.

Today's advice is very simple. Make sure the memory is rated for your machine's bus speed, then buy as much as you can afford to stuff in your machine.

DDR3 RAM is beginning to appear. Right now its extra expense over DDR2 is not worth paying, for all but extremely specialized needs. It is almost always far more useful to have 4GB of reasonably fast RAM, than 2GB of very fast RAM, in your machine.

For more technical stuff on memory architectures, see The Ultimate Memory Guide maintained by Kingston Technologies.

3.2. How To Pick Your Processor

Right now (late 2007), the chips to consider for running Unix are the the 64-bit AMD Opteron or its Intel equivalents. We've basically passed the point at which 32-bit chips are interesting for new desktop systems; choosing one won't save you more than small change any more. The transition isn't complete in laptops yet, which are mostly still using 32-bit chips to minimize their power draw.

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. In the last few years I've become a big fan of the AMD line. They used to be faster, cheaper, and better-designed than Intel processors; today Intel has clawed back the speed advantage, but AMD chips still deliver more performance than you're likely to be able to use and do it with lower power dissipation (thus, less noise and heat).

Many CPUs now are multi-core — that is, they have multiple CPUs on a single chip. This is very useful for doing something compute intensive (re-encoding video, compressing large archives, etc.) in the background and still having a responsive system for other work at the same time. At current prices, a dual-core CPU makes good sense for most desktop systems. If you are creating a server or have specialized computing needs you expect to be very CPU-intensive, quad-core is worth considering. Only at the very low end (sub US$50 CPUs) do single-core CPUs still make sense on desktop machines.

Mainstream desktop CPUs now use one of two sockets: LGA 775 (Intel) and AM2 (AMD). Buying a system that uses one of these stands more chance of allowing a useful CPU upgrade to extend its useful life than systems using other less common sockets.

Current CPUs are much faster than those of just a few years ago. As a result, unless your needs are highly specialized, spending more than about US$200 on a desktop CPU is hard to justify. Putting extra budget into more RAM or a faster disk subsystem will most likely result in greater benefit to most users.

3.3. One Disk or Two?

I usually build with two disks — one "system" disk and one "home" disk. There are two good reasons to do this that have nothing to do with the extra capacity. One of them is the performance advantage of being able to interleave commands to different physical spindles that we'll explain a bit later in the section on disks. The other is that I am quite a bit less likely to lose two disks at once than I am to trash a single one.

Let's suppose you have a fatal disk crash. If you have only one disk, goodbye Charlie. If you have two, maybe the crashed one was your system disk, in which case you can buy another and mess around with a new Linux installation knowing your personal files are safe. Or maybe it was your home disk; in that case, you can still run and do recovery stuff and basic Net communications until you can buy another home disk and restore it from backups (you did keep backups, right?).

Given today's high capacity drives, another way to use two disks well is to set them up as a RAID1 (mirrored) array. This can be done in software or with a hardware RAID controller. This way if either of the two drives fail, the system will continue to function, no data is lost, and upon replacing the failed drive, the array can be rebuilt from the remaining working drive. Hard drives are consumable media, they do fail, so this approach (as well as good backups) is well worth considering.

Your performance-tuning choice is SATA versus SCSI (the older IDE and EIDE buses are now obsolete). We'll have more to say about that in Section 4.

3.4. Getting Down to Cases

I used to say that cases are just bent metal, and that it doesn't much matter who makes those. Unfortunately, this isn't true any more. Processors run so hot these days that fans and airflow are a serious concern. They need to be well designed for proper airflow throughout.

Look for the following quality features:

• Aluminum rather than steel. It's lighter and conducts heat better.

• Unobstructed air intake with at least one fan each (in addition to the power supply and processor fans)

• No sharp metal edges. You don't want to shred your hands when you're tinkering with things.

• There shouldn't be any hot spots (poor air flow).

• Sturdy card clips. Some poorly-designed cases allow cards to wiggle out of their slots under normal vibration.

• Effective and easy to use mechanisms for attaching hard drives, CD-ROM, CD-R/W, DVDs, etc.

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 cost difference will be trivial even if they do. 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, 300W power supply, half-size motherboard, on-board SATA and 4GB of RAM sockets, and half-sized expansion cards, will fit into a baby or midsized 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.

3.5. Power Supplies and Fans

A lot of people treat power supplies as a commodity, so many interchangeable silver bricks. We know better — cheap power supplies go bad, and when they go bad they have a nasty habit of taking out the delicate electronics they're feeding. Also, the power supply tends to be the noisiest component in your system.

Give preference to supplies 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. On the other hand, processors and their cooling fans eat a lot more power than they used to.

The choice is generally between 200W and 300W. After some years of deprecating 300W-and-up supplies as overkill, I'm now persuaded it's time to go back to them; a modern processor can consume 50-75W by itself, and for the newer dual-processor board the power supply needs to be rated 450W or up.

Processors on modern motherboards run hot enough that all vendors have gone to embedded temperature sensors and variable-speed thermostat-controlled fans, out sheer self-defense (this used to be a high-end only feature).

To cut noise, look for 120mm fans rather than the old-style 80mm muffin fans. These can move the same amount of air per minute rotating at a lower tip speed, which means less vortex formation and less noise. These are now becoming standard even on cheap white-box hardware.

In garden-variety tower cases there often isn't enough airflow to cool all components effectively with a single fan, even going at full speed. And the single fan in the power supply was basically designed to cool the power supply, not the components in the case. This is why processors and some graphics cards have their own fans now.

A few years ago PCs often had two or more case fans in addition to the power-supply fan. This made sense in the era of 80mm fans and lots of expansion cards obstructing the airflow, but it was noisy. Nowadays, with sound and graphics and Ethernet integrated onto motherboards, expansion cards are much less common (and processors carry their own mini-fans). Thus, today's standard is to mount one 120mm fan, usually low and forward just beneath the disk-drive stack. This is much quieter, like by a factor of three or four.

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 cause lots of turbulent airflow, the fan produces lots of noise. One brand of fans that is much more silent than most others even if going at full throttle is Papst.

3.6. 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. 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 (even the "name" vendors use lightly customized versions of these). 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.

Some 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.

(I used to have "Voltage, temperature and fan speed monitoring hardware." on this list. But processors run so hot nowadays that all current motherboards have it.)

The dominant form factor is still ATX. Intel tried to replace it with a new standard called BTX in late 2004-2005, but failed; the proposal was effectively withdrawn in 2006. In January 2007 AMD announced a DTX specification for small-form-factor PCs; it remains to be seen whether this will gain any traction.

3.7. Monitor and Video

The largest user-visible change since the last major update of this guide is that the CRT (cathode-ray tube) is dead. The manufacturers shut down their production lines in late 2004; the remaining CRTs out there are old stock that's been sitting in warehouses. The only reason to buy one since then has been to get high-end resolution at a price lower than the insanely expensive high-end flatscreens; with 1920x1440 flatscreens now (late 2007) becoming generally available at reasonable prives even that reason is gone. It's all flatscreens now, baby.

On flatscreens, only two statistics matter; pixel size and response time. The biggest functional drawback of flatscreens relative to CRTs is that they refresh more slowly, because cheical reactions in a flatscreen pixel take longer than remodulating a flying electron beam. You'll never notice this during ordinary desktop use, but it can cause streakiness and artifacts when you're playing games or viewing movies. If you're going to do that a lot, the price premium for a flatscreen with better response time may be worth it.

Next, buy your card (if you have to; see next paragraph). This used to be complicated, with issues like matching the video bandwidths of the card and the CRT, and the amount of display memory. Now (unless you are a gamer or have similarly extreme 3D acceleration requirements) it's simple; all cards have enough display memory for every resolution in use, and the issues are software (does it have an open-source driver, and do you care?)

It's actually fairly likely you'll never buy a video card again. Very capable graphics chips are routinely integrated onto motherboards now; unless you're a gamer or somebody else who absolutely must have the latest wheeze in 3D acceleration, they'll be good enough. Even this is not a stable situation, as 3D acceleration is commoditizing too.

I used to carry a lot of material on different video standards, interlacing, and flicker. That stuff is all obsolete now.

Here's what to look for on the monitor spec sheet:

• Screen size and format. Usually measured in diagonal inches. Many screens are now in a "widescreen" format (16:10 ratio of width:height) rather than the older 5:4 or 4:3 ratios common for CRTs and older flat panel screens. A "19 inch" widescreen monitor generally has considerably fewer pixels than a "19 inch" 5:4 ratio one.

• Screen resolution. 1280x1024 is now low end on the desktop. Seventeen inch 1280x1024 screens are the bargain basement now, some manufacturers already switching production to 19 inch widescreen 1440x900 screens instead. In late 2007, the cost difference between such screens and 20 inch 1680x1050 screens is very small, making the 20 inch screens a better choice unless funds (or desktop space!) are very tight.

• 5ms or lower response time. 3ms is better. There is some marketing-speak going on in the way the response time is specified (grey to gray rather than black to white) but since most manufacturers do it this way these times are usually comparable between different manufacturers screens.

• Does it have a tilt-and-swivel base? Adequate controls, including both horizontal and vertical size and horizontal and vertical centering? A color-temperature control is a plus; the last is particularly important if you compose graphics on screen for hardcopy from a printer.

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 significant quality variation (even in "premium" monitor brands) even among monitors of the same make and model.

3.8. DVD Drives

DVD drives have two main uses in computer systems: playback of video DVDs, and use for data storage (either installation media or backups, or even as a primary drive in a few specialized systems).

3.9. Sound Cards and Speakers

You can't buy a really bad sound card any more. Even low-end sound cards or the sound chips embedded in a lot of PC motherboards these days support support all these features:

• 16-bit sampling (for 65536 dynamic levels rather than 256).

• Mono and stereo support.

• Full-duplex mode.

• Sampling rate of 44.1KHz (CD-quality).

• MIDI interface via a standard 15-pin D-shell connector.

• RCA output jacks for headphones or speakers.

• A microphone jack for sound input.

If you are interested in multi-track digital recording, two particularly good choices are the M-Audio Delta, or RME Hammerfall series of cards. Decent (and lower cost!) two-channel cards for more normal use are those using the ICE1712 (Envy24) and ICE1724 (Envy24HT) audio chips. For normal users, though, almost anything will work fine!

A rather comprehensive list of sound cards and chips supported by the ALSA project, which is the main way sound cards are supported under Linux, can be found at ALSA Sound Card Matrix.

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. Usually you'll want separately-powered speakers. If appropriate for your listening habits, a pair of decent headphones will get you better quality sound for the money compared to speakers.

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 games with sound even if your sound board won't accept the audio cable connector.

3.10. Modems

Demand for (dialup telephone) modems is dropping as more and more people get broadband Internet through DSL and cable. This section still has as much detail as it does only because (a) there are people out beyond the exurbs who can't get broadband, and (b) there are one or two remaining traps for the unwary.

The modem market has stabilized and standardized. If you can spend $59, get a U.S. Robotics V.92 USB external. You can then know that you've got the best and skip the rest of this section. If you really must economize, spend $39 for the internal-card version (but you'll probably regret the $20 first time you have to do diagnostics).

If you live somewhere with really bad telephone lines, the U.S. Robotics V.92 Business Modem may be truly "the best" for your needs, though it is about four times the price of the U.S. Robotics V.92 USB external, which is marketed for home use. See the U.S. Robotics web site for current product numbers and more detailed specifications.

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 all modems do a nominal 56kbps —V.90 and V.92 plus V.29 or V.17 fax transmission and reception (over plain old phone lines you won't get more than 53K of that). 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.

3.11. 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.

(There used to be a problem with "GDI printers" and "WinPrinters" that only worked with Windows —they required special drivers that took over your CPU to do image processing, These were such a bad idea that they have basically disappeared off the market.)

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 Windows 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. At today's prices, paying the small premium 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 — if you can find one. Inexpensive ink-jets and lasers have almost driven them off the market.

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...)

Nowadays, a lot of printers are moving away from parallel-port interfaces to USB. This is a good idea, because USB devices announce themselves to the host computer and can be automatically configured. Parallel ports (and serial ports for that matter) are becoming obsolete. Many new PC motherboards no longer include them.

Many printers (even some sub $100 models) now come with a network (10/100 Ethernet) interface. This make sharing them trivial, and also avoids having to leave a desktop PC powered on so others (using notebooks perhaps) can print to your printer. Therefore, such printers are worth considering in many networked environments, including home networks.

In the near future, new motherboards may stop including parallel and serial ports altogether. That's another good reason to go with a USB- or Ethernet-capable printer.

3.12. Power Protection

I strongly recommend that you buy a UPS to protect your hardware and data. MOV-filtered power bars make nice fuses (they're cheap to replace), but they're not enough. I have written a UPS HOWTO that provides more complete coverage of what used to be in this section.

3.13. 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 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 well 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.

4.1. First, add more memory

Max out your memory. Having lots of free memory will improve your virtual-memory performance (and Unix takes advantage of extra memory more effectively than Windows does). Fortunately, with RAM as cheap as it is now, a gigabyte or three is unlikely to bust your budget even if you're economizing.

4.2. Bus and Disk speeds

Most people think of the processor as the most important choice in specifying any kind of personal-computer system. But for typical job loads under Linux, the processor type is nearly a red herring — it's far more important to specify a capable system bus and disk I/O subsystem. If you don't believe this, you may find it enlightening to keep top(1) running for a while as you use your machine. Notice how seldom the CPU idle percentage drops below 90%!

It's true that after people upgrade their motherboards they often do report a throughput increase. But this is often more due to other changes that go with the processor upgrade, such as improved cache memory or an increase in the clocking speed of the system's front-side bus (enabling data to get in and out of the processor faster).

If you're buying for Linux on a fixed budget, it makes sense to trade away some excess processor clocks to get a faster bus and disk subsystem. If you're building a monster hot-rod, go ahead and buy that fastest available processor — but once you've gotten past that gearhead desire for big numbers, pay careful attention to bus speeds and your disk subsystem, because that's where you'll get the serious performance wins. The gap between processor speed and I/O subsystem throughput has only widened in the last five years.

How does it translate into a recipe in 2007? Like this; if you're building a hot rod,

• Do buy a machine with the fastest available "front-side" (e.g. processor-to-memory) bus.

• Do get a high-speed SCSI controller and the fastest SCSI disks you can get your hands on.

If you're economizing, you can back down on these. But in trading away SCSI for SATA your reliability (expected time before failure) will drop. We'll cover this in more detail in the next section.

4.3. Disk Wars: SATA vs. SCSI

For the fastest disks you can find, pay close attention to average seek and latency time. The former is an average time required to seek to any track; the latter is the maximum time required for any sector on a track to come under the heads, and is a function of the disk's rotation speed.

Of these, average seek time is much more important. When you're running Linux or any other virtual-memory operating system, a one millisecond faster seek time can make a really substantial difference in system throughput. Back when PC processors were slow enough for the comparison to be possible (and I was running System V Unix), it was easily worth as much as a 30MHz increment in processor speed. Today the corresponding figure would probably be as much as 300MHz!

The manufacturers themselves avoid running up seek time on the larger-capacity drives by stacking platters vertically rather than increasing the platter size. Thus, seek time (which is proportional to the platter radius and head-motion speed) tends to be constant across different capacities in the same product line. This is good because it means you don't have to worry about a capacity-vs.-speed tradeoff.

Disks of less than 40GB 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 9GB unit platters, which are simply being stacked 2N per spindle to produce ranges of drives with roughly 18GB increments of capacity.

Average drive latency is inversely proportional to the disk's rotational speed. For years, most disks spun at 3600 rpm; most disks now spin at 7,200 or 10,000rpm, and high-end disks at 15,000 rpm. These fast-spin disks run extremely hot; cooling is becoming a critical constraint in drive design.

Another basic decision is SATA vs. SCSI (the older IDE and EIDE buses are obsolete). 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 a SATA 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. You can have at most four SATA devices on a single controller. SCSI permits up to 7 (15 for wide SCSI).

Admittedly, the case for SCSI has eroded a bit since 2001; the new generation of SATA drives is very fast, and controller cards now normally feature a channel per drive and DMA (Direct Memory Access), so that some of of the multi-user contention problems that used to dog IDE have diminished. At 10KRPM and below, SATA is as good as SCSI now (a painful admission for an old-time IDE-hater like me), but at the 15KRPM high end SCSI still rules.

Of course, SATA is cheaper. Many motherboards have SATA right on board now; if not, you'll pay maybe $15 for a SATA 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.

 +--------+   +-------+     +-----------+     +-------+
 |Internal|---|Adaptec|-----|New Quantum|-----|UMAX   |
 |Devices |   |1542CF |  ^  |  Disk     |  ^  |Scanner|
 +--------+   +-------+  |  +-----------+  |  +-------+ 
                         |                 |
                    New $16 cable   Cable that came
                                      with scanner.

4.4. Tuning Your I/O Subsystem

(This section comes to us courtesy of Perry The Cynic, <perry@sutr.cynic.org>; it was written in 1998. My own experience agrees pretty completely with his. I have revised the numbers in it since to reflect more recent developments.)

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. (Beware that if you mix an IDE CD-ROM with SCSI drives under Linux, you'll have to run a SCSI emulation layer that is a bit flaky.)

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" (this is a specialty physical layer 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, 10000, or 15000 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 Fujitsu MAM) runs 3.9ms/15000rpm.

Fast, big drives are expensive. Really big drives are very expensive. Really fast drives are pretty expensive. 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 (early 2004), it's about at 36GB drives, 6ms, 10000rpm, ultra2 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 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 :-).

SCSI performance can sometimes be improved by setting the ID of the most frequently used disk drive as high as possible. The SCSI priority pecking order is such that devices with higher ID's get first crack at the bus when arbitration occurs during the selection phase.

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 note 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. :-)

8.1. 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 good way to cope with this is to configure your system on paper, get a couple of initial estimates, then set a trigger price, below the lowest one, at what you're willing to pay. Then watch and wait. When the configuration cost hits your trigger price, place your order.

The advantage of this method is that it requires you to settle in your mind, well in advance, what you're willing to pay for what you're getting. That way, you'll buy at the earliest time you should, and won't stress too much out afterwards as it depreciates.

Before you shop, do your homework. Publications like "Computer Shopper" (and their web site at http://www.computershopper.com) are invaluable for helping you get a feel for prices and what clonemakers are doing. Another excellent site is ComputerESP.

8.2. 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.

8.3. 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. If you still have one in your neighborhood, you're lucky. I do, as it happens, but that is now unusual; the only place you normally find diaspora Chinese and Indians selling cheap PCs over the counter anymore is at computer fairs. (Usually they're doing it to publicize an Internet/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.

8.4. Mail Order

Internet buying makes a lot of sense today for anyone with more technical savvy than J. Random Luser in a suit. Even from no-name vendors, 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 Internet/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 parts is not likely to save you money over dealing with the Internet/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.

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 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.)

8.5. Computer Superstores

Big chain superstores like CompUSA and Circuit City give you a reasonable alternative to the Web. And there are good reasons to explore it — these stores buy and sell at volumes that allow them to offer prices not far above the Web. (They make back a lot of their margin on computer games and small accessories like mouse pads, cables, and floppy disks.)

Note, however: Avoid Best Buy. Horror stories about them are legion — predatory salescritters, incompetent service, routine bait-and-switch tactics.

One thing you should not buy remotely if you can avoid it is a monitor. Monitors are subject to significant quality variations even within the same make and model. Flatscreens haver this [roblem less than CRTs did, but you don't want a flatscreen with dead pixels. 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 system back to the vendor. 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 two-button mice that you can't use. Use caution and check your system manifest.

But if you shop carefully and don't fall for one of their name-brand "prestige" systems, you can get prices comparable to Internet/mail-order with the comfort of knowing there's a trouble desk you can drive back to in a pinch. (Also, you can see your monitor before you buy!)

8.6. 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 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.

9.1. 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.

9.2. 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.

9.3. 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:

• 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.

10.1. Tricks and Traps in Warranties

Reading 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-Windows 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.

10.2. Special Questions to Ask Web/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!

10.3. 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.

10.4. Which Clone Vendors to Talk To