Beowulf - Single Board Computers? (long)
Robert G. Brown
rgb@phy.duke.edu
Tue, 22 Sep 1998 09:17:52 -0400
On Mon, 21 Sep 1998, Bob Drzyzgula wrote:
> On Mon, Sep 21, 1998 at 01:49:37PM -0400, Walter B. Ligon III wrote:
> > --------
> >
> > The CESDIS Beowulf group built their current machine (ecgtheow) from
> > dual PII motherboards with built-in SCSI and 100baseT. You do NOT need to run
> > 110VAC to each motherboard, you run +/-5 and +/-12 (or something like that.
> > One could easily design a power supply to operate N motherboards. People
> > have been designing powersupplies for years.
>
> 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).
> The 3.3V is optional but probably required if you're
> using 3.3V SDRAM. The standby (several hundred mA) has
> to be delivered in advance of the power-up, I believe,
> unless you disable the soft power on/off somehow, so you
> have to wire it up separate from the regular +5V feed.
> The power supply isn't supposed to deliver power other
> than the 5V standby unless it gets the PW_ON signal from
> the motherboard, so you need some sort of per-motherboard
> logic control on the power feed lines.
That standby/powerup power can be a real bear. I got one and only one
Intel DK440LX motherboard-based system (from SWT) and went through three
power supplies befor finding one that could deliver the power surge
required to get the system to actually turn itself on. Immensely
frustrating to have a perfectly "good" power supply, 3,5,12V as promised
on every line, and have NOTHING happen when you press the "on" button.
Same damn thing, actually, happens when one tries to turn on e.g. a TV
attached to a voltage converter in a car or van (I have small kids and a
TV/VCR in the back of our van makes long trips bearable:-). Unless the
converter can deliver something like 500W peak (a pretty hefty current,
if you think about it) it isn't enough to get the TV over its initial
inductive surge and into its 100W stable operation zone.
This is a dark side of the ATX spec, as it means that your system is
especially vulnerable to voltage spikes or surges -- even when the
system is "off" there is a direct current path from the wall (well OK,
through the transformer do don't quibble about "direct") into the
motherboard. Anybody know why they did this? Was there something "bad"
about having one switch deliver power to the whole system?
I agree that this mildly complicates converting a 2-drawer filing
cabinet into a beowulf (power supply, network, head motherboard in the
top drawer, 8 dual PII's in the bottom drawer, pullthrough fans in front
and rear of each drawer:-). However, a "big" power supply (with a nice,
big filter capacitor:-) is actually safer and cooler at operating load
than a bunch of smaller ones (and isn't necessarily physically all that
much bigger). The biggest problem with the ATX scenario isn't getting
power in -- it is building all the auxiliary wiring harness and the
physical installation. You have to be able to control the on button,
the reset button, it's nice to pipe all the disk activity indicators
into a panel of LED's, there is the physical placement and connection of
the hard disk (if any)... a lot of hassle if you are truly building one
of these yourself from ATX parts. That's why I'd recommend going the
AltaTech or Paralogic route (unless you are a hobbyist converting a file
cabinet on a shoestring budget or for fun) if you do ATX. They handle
all of this for you at a very modest markup and in an attractive case.
I must confess, though, that I also like Walter's laptop motherboard
concept as well. Simple power supply (and lower power consumption).
Easily colocatable hard drives in a tiny form factor. Limited but
adequate networking, I think (at most two interfaces/motherboard, and I
don't know what the PCMCIA bandwidth latency specs are -- anybody? Is
there a GBE or Myrinet PCMCIA card? Can the interface actually handle
even 100BT with a decent interpacket latency?). However perusing
Computer Shopper looking for laptop motherboards turns up every AT and
ATX form factor under the sun, and no laptops motherboards. How would
one go about finding them? Surely there are standard manufacturers --
or does every laptop maker roll their own?
rgb
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb@phy.duke.edu