Beowulf - Single Board Computers? (long)

[Donald Becker]

As you might guess, we've looked at packaging.  Our first cluster was
repackaged into a 2'x3' rack with one AT power supply per four processor
shelf.

You might notice that our subsequent clusters are space-wasting boxes on
shelves.  The power supplies were never the problem, it was the physical
mounting that wasn't robust or easy. 

On Mon, 21 Sep 1998, Bob Drzyzgula wrote:

> Well, yes. However, the ATX power supply harnass is fairly
> complex as these things go. I believe that it requires at
> many as six voltages (+/-5V, +/-12V 3.3V and +5V standby).

I don't think it requires -5V, but it does add 3.3V to the list.  The other
voltages are required for ISA bus compatibility, although -12V is used only
for serial ports.

> The 3.3V is optional but probably required if you're
> using 3.3V SDRAM. The standby (several hundred mA)  has

The standby current isn't well specified.  That's why it's a unpredictable
if wake-on-LAN will work on a randomly configured system.

> to be delivered in advance of the power-up, I believe,
> unless you disable the soft power on/off somehow, so you

You can jumper the standby voltage to the on switch line.  Two of our
motherboard types had undocumented headers to do just that.  I found the
first header while looking for the easiest place to solder a jumper wire.
Without the soft-on bypass these motherboards would leave the machine
powered down after a power failures.

In theory you can share a ATX single power this way, but I haven't tried it
myself.  Mostly because of the lack of connectors.  Wiring up our first
cluster was painful and expensive, since contacts and connector shells that
cost pennies in Pacific rim countries were expensive and had long lead
times here.


Donald Becker					  becker@cesdis.gsfc.nasa.gov
USRA-CESDIS, Center of Excellence in Space Data and Information Sciences.
Code 930.5, Goddard Space Flight Center,  Greenbelt, MD.  20771
301-286-0882	     http://cesdis.gsfc.nasa.gov/people/becker/whoiam.html