(trans)portable Beowulf

[Jim Lux]

I am contemplating building a (trans)Portable Beowulf for field reduction
and visualization of measurement data. It's a fairly computationally
intensive problem basically involving creating an FEM model that replicates
observed measurements, but, fortunately, is fairly coarsegrained, so
interprocessor comm requirements aren't dominating (yet).

The basic requirement is that the beast be luggable/shippable/etc. and use
12V battery power. The initial design cut uses diskless nodes (i.e.
motherboard,video, and NIC) packaged into a custom enclosure of suitable
size with appropriate cooling and environmental controls. The power will
eventually supplied by DC/DC converters.

Here are the questions:

Real power consumption - how much current on each voltage is being drawn by
a typical motherboard? Without a disk, there probably isn't much draw on the
+12, -12, or -5 lines. So, there would be significant power on the 5V, the
3.3V, and the 5VSB lines. Anybody ever hooked an ammeter on the PS?  What
about "power save" modes supported by some motherboards/CPUs?  Wake on LAN?
The typical usage scenario is sort of "crunch and visualize for a few tens
of minutes... go collect some data for a few tens of minutes .. repeat".
It'd be nice to "suspend" the machine while we're doing something else.. it
makes cooling and battery life better.

What sort of options are viable for the "head node", assuming the nodes are
booting from it and using its disk? A laptop with a big disk? (The idea
being that you can "carry-on" the laptop, and check the rest as baggage (or
throw it in the back of the truck, etc.).)   Or, should I design in a
"ruggedized" head node with a big disk, etc., and then talk to it with a
relatively skinny laptop running an Xserver or ssh(or a Palm Pilot <grin>).

Once we are set up and configured in the comfort of the lab at home, where
the monitor, keyboard and mouse on the slave nodes are nice to have, how
important is it to have a monitor/keyboard/mouse in the field? It might be
several days before you could get back home. How much could one
realistically fix? If the node is dead, and you don't have spare parts,
having a monitor doesn't help much.  Are there failures where the node
couldn't boot due to some misconfiguration (which COULD be fixed with a
monitor/kbd, etc)?.


Jim Lux
Spacecraft Telecommunications Section (336)
Jet Propulsion Laboratory
4800 Oak Grove Road, Mail Stop 161-213
Pasadena CA 91109
Tel: (818)354-2075.  Fax:(818)393-6875