Atomic I/O letters column #122Originally published 2011, in Atomic: Maximum Power Computing
Reprinted here October 18, 2011 Last modified 03-Dec-2011.
Is it bad to make computer hardware out of Lego?
I don't mean CPUs, I mean cases and drive-holders and such. I've seen some fancy all-Lego cases on the Web, but all I really want to do, to start with at least, is make a sort of semi-permanent case for a hard drive plugged into a USB-to-SATA adapter. I've got a big tub of Lego from when I was a kid, and mechanically at least it seems really suitable for the job. You can make pretty precise shapes with it, and it looks better than just sitting drives on the table with cardboard between them or something.
All-plastic computer enclosures of any kind are illegal to sell in many countries, because they do indeed fail RF-interference tests. It's legal to make your own plastic case, though, and several companies have sold plastic PC cases to customers who may, or may not, then go on to install components in them. Who can say?
(See also the complicated devices offered in many tobacconists, for the smoking of tobacco or "herbs".)
Realistically, this is seldom a problem. RF noise from PC components is unlikely to be powerful enough to do anything beyond fuzz up AM radio reception within spitting distance.
The RF noise gets out because plastic is non-conductive and can't be earthed, which is also why it accumulates static electricity. That's more likely to be the real problem you'll face. An undetectably small static spark can damage computer hardware, and any anonymous piece of plastic can build up such a charge just by sitting in a dry breeze.
Being realistic again, though, I wouldn't worry much about this either. Static charge on your own body is the usual cause of hardware-zapping, and you can get rid of that tolerably well by just touching metal at the same potential as the hardware before you touch the hardware. For working on computers, that means touching the chassis metal somewhere; for drives by themselves, the metal plate on top of the drive is connected to the ground pins on the power connector. Even if you're running a drive from the ungrounded plugpack of a USB connection kit, touching the metal on top will still bring you to the same potential as the drive.
So feel free to follow in the footsteps of Google and build computer components out of Lego. And cases, of course, are only the beginning!
This will sound like a stupid question, but which brand are the best hard drives?
I'm mainly asking about reliability. I don't care if Photoshop takes 5 more seconds to load, but I really do care about drives dying. Surely you (by which I mean you, Dan, and also Atomic in general) have been playing with PC drives for long enough to get a general idea? Is there any real reliability trend between manufacturers?
It's generally impossible to tell how reliable most computer components are, until it's too late.
There are some exceptions to this rule. The high failure rate of Xbox 360s, for instance, or IBM's legendary "Deathstar" Deskstar hard drives, from when 100Gb was a lot of disk space. Usually, though, you can't get a reasonable idea about computer-part reliability until long after the parts are obsolete.
Statistics that're too late to save you from buying, say, those G84 and G86 8000-series Nvidia GPUs that had dodgy underfill in the chip package, could still be useful if they painted the picture you're looking for. If, say, Nvidia GPUs seemed consistently less reliable than ATI/AMD ones, or if Seagate drives seemed better than Samsungs. But there's no consistent picture there, either. Technology changes so fast that IBM's run of lousy hard drives had no effect on the reliability of their (now Hitachi, soon to be Western Digital) descendants, and all of those red-ringed Xbox 360s very probably say nothing at all about the reliability of the next Microsoft console.
All sorts of technological objects have "Mean Time Between Failures" numbers; hard-drive MTBFs are often outrageously high, in the hundreds of thousands of hours. Those numbers are made by testing, say, 1000 drives for a thousand hours; if two of them die over that period, then the simplest kind of MTBF calculation says the MTBF should be 500,000 hours, about 57 years. In reality, it's unlikely that any of the drives will live for more than ten years, even if they're not running all day, every day.
(SSDs have even higher MTBFs than moving-parts drives, by the way. Long-term use of memory cards and industrial Flash-RAM devices tells us that SSDs ought to last substantially longer than hard drives, in most applications, but real-world numbers haven't been nearly that good, though many people don't care. In any case, you definitely shouldn't expect any SSD to actually work for the two million hours listed as MTBF on the spec sheet; that's 228 years.)
All this doesn't stop people forming strong opinions about particular brands. If someone's had two Western Digital drives fail in a row, or two Samsungs, or two Seagates, they'll often now swear by whatever brand hasn't yet treated them so badly. In these cases I recommend the prompt administration of Darrell Huff's classic slim volume, "How To Lie With Statistics". It applies as well to hard drives in 2011 as it did to tooth-powder ads in 1954.
I have not laughed so hard in a long while as when I just read the first paragraph of section 9.1 of the comp.compression FAQ.
Especially when I hit "applicable recursively".
The things some folks can convince themselves of...
Only later did I note in the preface of the page:
"This topic has generated and is still generating the greatest volume of news in the history of comp.compression."
Yeah, I love that stuff. Compression programs that can always make any data smaller are to mathematicians and computer scientists as perpetual-motion machines are to physicists.
The best part is that the theory that makes clear that this is impossible, called the "pigeonhole principle", can be explained even to people (like me...) who have no knowledge of advanced mathematics at all. Only if someone's really determined to say "you know, science doesn't know everything, FTL and antigravity could be possible, and so could fitting 100 things in 99 pigeonholes without any pigeonhole having more than one thing in it" will they then be fooled by any of the endless series of bogus magical data-squishers.
Often, the people who write such programs are just cranks. When a dedicated scam artist goes at it, the results can be much more impressive. "Infima" was pretty awesome; just a bunch of other people's compression programs stuck nose to tail, plus a dodge to make sure decompressed files tested as bit-identical to the originals, because the software overwrote the originals with a copies of the compressed versions when the hapless user wasn't looking!
Those USB drink warmer things didn't work, because USB can only deliver 2.5 watts. Now we've got USB 3, though, and apparently that can deliver more power. So will it now be practical to have USB soldering irons and USB toaster ovens?
Officially, as you say, USB versions 1 and 2 can supply no more than half an amp (500 milliamps) at five volts from a powered port (a "root port" on a controller, or a port on a powered hub). And, again officially, USB 3.0 can indeed deliver more juice.
But nothing is simple.
A "Battery Charging Specification", separate from USB 3, was added to the USB spec in 2009. A powered USB port meeting this spec should be able to deliver as much as 1.5 amps in the two USB 1 speed modes, and if there's no data transfer at all, there's no ceiling current beyond the carrying capacity of the connectors. Normal USB A connectors are only specced for 1.5 amps, but manufacturers are allowed to use beefier ones if they like.
(Remember that current actually delivered into a given load depends on the load's resistance; this is why you can grab the terminals of a car battery without being vaporised by hundreds of amps. Little LED fan doohickeys and awful $4 electric razors will work the same from a "high power" USB socket as they did from a normal one.)
You still shouldn't expect to be able to, say, run a forty-watt laptop from 5V USB power; even with zero loss that'd be eight amps, which'll make USB-cable conductors rather toasty. I could see a netbook running from nothing but USB power, though. And you actually can already buy a USB soldering iron, which runs from one of those double-headed Y cables that suck power from two USB ports, and may actually work, sorta. Proper soldering pencils with power draw under 20 watts are mainstream products, so beefier USB could probably run one of those OK, too.
All of this specification discussion is of limited relevance, though, because ugly things happen when specifications meet the real world.
Take the abovementioned double-headed USB power cables, for instance, which are commonly used to run little external drives that won't quite spin up from 2.5 watts. The official USB spec doesn't allow those. Heck, it doesn't even allow extension cables.
There are a bunch of USB 1 and 2 specifications - minimum device voltage requirements, cable length and type restrictions - that are interpreted very loosely indeed by a lot of manufacturers. Some USB controllers can provide more power than they're meant to, some USB devices are touchier about technically-OK supply voltage and cable lengths, et cetera. If I were you, I wouldn't expect USB 3 to bring better compliance with the rules, especially by manufacturers of dirt-cheap USB peripherals.
I wouldn't be at all surprised if, to achieve maximum USB compatibility, a normal computer two years from now will need its motherboard USB 3-point-whatever-we're-up-to-by-then controller, plus an add-on USB 3.0 card, and an add-on USB 1/2 card. Since add-on USB 3 cards are already cheap and USB 2 cards are now practically free, this isn't likely to be a major problem unless your teeny computer or laptop in the year 2013 has nowhere for you to plug more cards into it.
You can be pretty confident that any USB 3 controller will be able to deliver the 900mA per port that's in the USB 3.0 spec. That's only 4.5 watts, but it's better than 2.5W.