[Beowulf] Similar to a multi CPU machine?

[Robert G. Brown]

On Mon, 14 Nov 2005, Jake Thebault-Spieker wrote:

> I've heard of many of these(Scyld, OpenMosix, etc.) Since this is how it will
> work, will I be able to use these 6 machines as a workstation? I understand
> that the CPU power will be next to none, and I have read your book about
> overhead, and feasability and so forth. Having not done the calculations, I
> have a pretty good feeling that my computers will not meet the standards for
> them to actually be useful. I'm a high school student, and would like a
> multi-CPU machine to play w/, but I don't want to spend the money. Thanks
> again.

To use the six machines "as a workstation" will probably require
OpenMosix -- that permits you to login to one machine's console and just
run jobs.  Those jobs will then automigrate onto nodes, with network
socket connections established to pipe e.g. stdin, stdout, open file I/O
back to the master node transparently.  For embarrassingly parallel
tasks -- independent runs of some background job -- this will probably
be "reasonably" efficient.  For graphics intensive tasks, for tasks with
a lot of I/O, it probably will be less efficient or even INefficient --
you add network overhead on top of the already nontrivial I/O overhead
and a different form of virtual parallelism might be called for that
permits the usage of local disk resources (for example) rather than
always writing to a virtualized local disk that then is socket-forwarded
back to the master host and turned in to a real disk write at that end,
especially if the master is running many copies of a similar/identical
task so that its disk AND network subsystems are constantly colliding
and nodes are having to wait in line (blocking the calling application
in the meantime).

So sure, give OM a try.  If nothing else, it will be a learning
experience for you -- don't worry too much about whether or not the
computers will be "useful" per se or whether you could do better to just
buy a new AMD-64 motherboard unless or until your task is worth
real money.  Otherwise "useful" is defined mostly by whether or not you
have fun with it, how much you learn from doing it.

     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 at phy.duke.edu