[Beowulf] portable clusters

[Jim Lux]

At 10:40 AM 11/28/2005, Andrew Piskorski wrote:
>On Mon, Nov 28, 2005 at 07:08:33AM -0800, Jim Lux wrote:
> > At 09:57 PM 11/27/2005, Greg Lindahl wrote:
> > >On Sun, Nov 27, 2005 at 09:32:45PM -0800, Jim Lux wrote:
> > >> Has anyone done a diskless cluster (i.e. boot the nodes off the network)
> > >> where the nodes run Linux but the headnode/server is a WinXP/Win2K box?
> > >>
> > >> Think in terms of booting off the drive in a laptop.
>
> > >Why does the Windows box need to be the head node? Just make it a
> > >dedicated data collection system, and have a normal Linux head node.
> >
> > Only want to have ONE disk in the system, and it needs to be in the 
> Windows
> > box to support the the data acq function.  Dual boot of the laptop isn't a
>
>Why?  That seems like an artificial restriction to me.  Are you doing
>someting like carrying your laptop around to conferences and such,
>booting borrowed desktops from CD to use as the (diskless) nodes, and
>you don't want to carry around a second laptop or other small machine
>to use as the Linux head node?

Aren't all design requirements artificial to a certain extent?<grin>

But basically, yes, the idea is to minimize the number of boxes.. One, the 
Windows laptop which talks to the hardware and feeds the cluster to do the 
data analysis, the other, the cluster box (we're not talking hundreds of 
nodes here.. probably more like 8-20)  The cluster box is envisioned to 
have two wires: power and network, and is generically, the size of a big 
breadbox or a suitcase(i.e. it would fit in the trunk of a car or could be 
checked as luggage...)

Ideally, the cluster has NO permanent storage within it (i.e. if you turn 
off the power, all inside is forgotten). Removable media could potentially 
solve the problem, but then you've added a third thing to carry around and 
potentially lose, as well as adding a connector that has to be dealt with. 
Having it totally diskless helps with some environmental requirements (you 
can drop it, shock it, or vibe it, and it's easier to deal with the dust 
ingress problem)...

The goal here is to have a credible concept to improve the "system" 
performance by adding on a computatational element to an existing portable 
system that is Windows based without requiring any hardware changes to the 
windows system, or without requiring significant software mods to the 
windows system (i.e. running some new program is ok, running a windows 
emulator is not).



James Lux, P.E.
Spacecraft Radio Frequency Subsystems Group
Flight Communications Systems Section
Jet Propulsion Laboratory, Mail Stop 161-213
4800 Oak Grove Drive
Pasadena CA 91109
tel: (818)354-2075
fax: (818)393-6875