Atomic I/O letters column #58
Originally published in Atomic: Maximum Power Computing Reprinted here June 2006.Last modified 16-Jan-2015.
Flashing lights
An idea occurred to me after reading the Atomic review of the Gigabyte i-RAM and noticing the 150Mb/s limit on SATA bandwidth. Why not use optical cable to connect devices? So I Googled it and found the speed of optical to be up to 10Gb/s! Googling further I found something Intel was doing.
So the idea ain't new, and there seem to be some developments... but what bugs me is that this technology already exists! I've got optical cable connecting my DVD player to my sound system!
So, just out of curiosity, why don't we already have optical connections?
Jason
Answer:
Copper may be running out of puff for super-fast CPU connections - motherboard makers have been doing tricky things
with circuit paths for many years now to equalise signal transit time - but
it's a long way from its limits in the storage-interface department.
People are using optical connections for storage and networking right now. Fibre Channel (FC) over optic fibre is in common use in the enterprise storage market.
The advantage of FC, though, is that it lets you hang a vast drive array or whole busy network off one slim cable, and run that cable quite a long way. If you're only doing home or small business computing tasks, you don't need that much bandwidth or range. Even if you set up a proper fast RAID array, one of the much cheaper fast SCSI interfaces, or even FireWire, will be fast enough. And your array probably won't be in a building a quarter mile down the road.
When you need to pack huge amounts of data into a long distance cable, optical's the way to go - it's hard to make a phone call today that isn't photons for some part of its trip. But for less demanding purposes, and for lower cost, copper still works just fine - as evidenced by the fact that a lot of the world's FC setups actually run the FC protocol over twisted pair copper cable!
Your home theatre system uses an optical cable for... well, for no terribly good reason, actually. The highest bandwidth that S/PDIF cable's ever going to have to carry is that of cinematic DTS audio, which only needs the same 1,411,200 data bits per second as uncompressed CD audio. DTS from DVDs is usually only 768 kilobits per second, and most DVDs only have AC-3 audio, which never takes up more than 448 kilobits per second.
All of these bit rates are pathetic even compared with ancient 10BaseT Ethernet, and well within the abilities of a very cheap electrical cable, as all the people who use electrical S/PDIF over a $4 RCA lead can testify.
Heavy metal
Cooling is one of the things that is constantly on my mind, lately I have been slowly designing a water-cooling kit.
One of the things I was considering was the use of mercury instead of water.
Mercury, as a metal, would have better heat carrying properties than water. It would be a kind of moving heatsink, as well as looking really funky and being really Atomican.
I have been told that mercury will expand hugely when heat is applied, so you would have to provide a huge amount of room in the tank for the mercury to expand and a powerful pump to push it around, due to the greater density.
What is stopping us from using a mercury based cooling kit? The plusses seem at the moment to be outweighing the negatives.
Nathanael
Answer:
What's stopping us?
Oh, just one or two minor things.
I've a great affection for Hg, as regular readers will know, but as a computer coolant it's a remarkably poor choice.
First up: Mercury has indeed been used as a coolant - in some old nuclear reactors. Modern liquid-metal-cooled reactors typically use something like sodium instead, because molten sodium spraying all over the place is actually not the biggest problem, if the reactor's cooling system's ruptured.
Plus, if you spill a drop of mercury on your nice aluminium case and don't notice, it will have eaten a hole through it and seeped into the carpet by the next morning. It'll do the same thing to any aluminium cooling jackets it passes through.
Mercury doesn't, however, actually expand very much when it gets hot. It's got a quite high coefficient of expansion compared with most metals, but you don't typically notice your car getting longer on hot days. We're still only talking an expansion of about 2% for mercury, from -30 to +70 degrees C. A mercury thermometer makes it look as if there's a lot of expansion going on, but the capillary tube the bottom bulb of mercury expands into is actually extremely thin - it looks thicker because of the magnifying effect of the curved glass over it. Mercury's used in thermometers because it's liquid over a usefully broad temperature range, and has quite linear expansion as its temperature changes.
You're not kidding about needing a powerful pump, either. Water weighs one gram per cubic centimetre; mercury weighs 13.5 grams per cc. That's 1.7 times the density of iron, and almost 1.2 times the density of lead.
Sub-reagent-grade mercury (more than good enough for use as coolant) costs around $US95 per half-kilo from scientific suppliers. You'd probably get a discount if you bought a litre of the stuff, but you'd still be paying at least $2700 Australian. If you could find an industrial supplier that'd sell you a standard "flask" of mercury then you could get 34.5 kilos (about 2556ml) for a mere $US850 (prices have spiked recently). But if you needed it shipped, that wouldn't be cheap either.
(Think of a number, multiply by 20. That's if you can find a shipping agent that'll touch mercury in the first place.)
The cheapskate option is salvaging mercury from old instruments that use it - old blood pressure meters are favourite, since you can in theory get more than a cubic inch (221 grams) of mercury out of just one of them. In my experience you don't get nearly that much from old sphygmomanometers, but even if you do get a cubic inch from each then you'd need about 50 of them to get a litre.
And then, if you ever wanted to dispose of the mercury legally, you might well have to pay more than you paid to buy it.
Why, you innocently ask, is that?
Because mercury's POISONOUS, that's why. It's a heavy metal that can be absorbed through the skin, lungs or digestive tract, and it accumulates in the body and is rather difficult to remove. It primarily affects the brain and liver, two organs to which most people are quite attached.
Metallic mercury emits vapour constantly at room temperature - on account of how it's, you know, molten metal - and readily breaks up into tiny droplets with large surface area which emit much more vapour.
Despite what the anti-amalgam-fillings quacks say, modest metallic mercury exposure is not, actually, a big deal. Which is good, because it's hard to avoid atmospheric mercury from the exhausts of the world's coal-fired power plants, much less avoid similar inhalation exposure if you're a kid sitting in a science classroom where someone broke a thermometer last year. And metallic mercury's a pussycat compared with some organic mercury compounds; one small drop of dimethylmercury will zip straight through a rubber glove and do you fatal damage in less than a minute. And it's organic mercury that you're exposed to by eating fish.
So if you overcame the engineering and financial challenges, you could assemble a mercury-based cooling system quite safely. Wear rubber protective gear, and do the job outside. But if the system wasn't well sealed, or if a hose popped off at any point, your computer room would instantly become a toxic waste clean-up site.
Hoover-induced failure
I have a generic Taiwanese MP3 player. I use it while I'm at work (cleaner). I have no problems with it until I start using the vacuum cleaner.
Whenever the MP3 player gets near the vacuum cleaner, the player restarts. I have no idea why.
It doesn't do it any other time (except when the battery is about to go flat, and I check this every time it happens).
So... any ideas why this might be happening?
Morris
Answer:
Vacuum cleaner motors, like a lot of cheap electric motors, emit a lot of RF noise. Vacuum cleaners, power drills,
electric shavers; all spark merrily inside, which is why you should never use a normal vacuum cleaner to suck flammable
vapour out of anything. Sparks create broadband RF noise, often way up into the microwave range - so motors can sometimes
screw up wireless networking connections, too.
Cheap MP3 players (and, I'm sure, some of the expensive ones too) have crummy shielding, and the RF noise induces current in their circuitry and freaks them out. You might notice similar symptoms if you left the player sitting next to a digital mobile phone when it chatted with the local cell tower - various gadgets, including some fancy LED flashlights, can be affected by that.
There's not a lot that can be done about this. Wrapping the player in the Unmedicated Schizophrenic's Friend (well, usually), aluminium foil, will only work if the foil is earthed. It'd get a (crummy) earth if it was in contact with your body, though, which might be enough. Otherwise, just keeping the thing away from the cleaner would be adequate - look like a dork and wear it on a neckstrap or headband or something.
Objective Performance Criteria?
I have a Philips CDR 600 CD recorder. It used to work fine, but now almost every time I try to record to a disc, the machine takes ages to do its verification bit and then flashes up "OPC ERROR".
Is this a disc problem? Is the machine telling me it's had enough and is retiring? Can it be fixed, or is it a replace and chuck the old one situation?
Pete
Now start licking.
Answer:
The error means the recorder's failing to do its power calibration
- the initial test it does on the disc to determine the write laser power it needs. And yes, it's probably symptomatic
of a larger problem; CD/DVD writer laser diodes do wear out eventually.
If you haven't recorded quite a lot of discs, though, the diode should still be OK - but there might be some mechanical problem, too. These things are built down to a price, not up to a standard.
It's possible that this error, or any number of other errors, may just be the result of fluff on the lens. One of those cleaning discs with the teeny brush sticking out of it ought to clean the lens quite well without requiring you to fish around inside the machine with a long cotton swab. Cleaning discs are unlikely to be well balanced, though, so you should not put them in high speed computer CD/DVD drives. Standalone recorders like yours ought not to spin that fast.
The best way to clean optical drive lenses is by taking the gadget apart, so you can actually look at the lens and clean it with a little brush, or whatever. This isn't actually very hard or dangerous (all you're likely to need is a couple of Phillips screwdrivers, and yes, you should unplug the recorder first...) but can still be daunting for beginners.
Note that just sticking an air duster straw in there can fix a dust problem, or make it worse. There's probably plenty of dust in there, which is perfectly harmless unless you stir it up and it goes somewhere important.
It's a (bigger) RAID!
In your "Cheap giant bit barrel" column, you said it is not possible to expand a software RAID 5 array without wiping the array.
Is it not possible to expand the array with Microsoft's "Diskpart" program?
Vincent
Yes, this is another one of those tools with which you can totally thork your PC.
Answer:
Nope; Diskpart can only expand basic volumes, and dynamic simple or spanned volumes.
Basic volumes can only be expanded onto contiguous space on the same disk, but that's what you get from a proper RAID controller; a hardware RAID array looks like one physical disk to the computer. Adding disks to the array makes it look as if that disk has magically expanded.
All proper RAID controllers let you add disks. Cheap ATA RAID controllers still don't. This is presumably because there's no real demand for that feature in the entry level pseudo-hardware (cheap controllers all do some of the RAID work in the driver) RAID controller market. Most people who use these controllers, including me, don't use the RAID features at all.
But fancier ATA RAID controllers let you add disks - 3Ware's 9000-series SATA controllers, for instance, have what they call "On-line Capacity Expansion (OCE)".
This is all very well, but Diskpart (like other partition-twiddling software) can't expand software RAID.
