Re: tlug: Re: multi-processor linux configuration ?

[Darren Cook <darren@example.com>]

>I got some email from a good friend of a good friend of mine, Andrej,
>who is using Linux to do parallel and distributed processing to solve
>real problems in biochemistry.

When I asked a similar question a month ago these were the links I was given:

http://www.linux-howto.com/LDP/HOWTO/Parallel-Processing-HOWTO.html
  (covers all the options)
http://garage.ecn.purdue.edu/~papers/
http://beowulf.gsfc.nasa.gov/    (NB. URL has changed recently)
http://legion.virginia.edu/

If you're writing the software, and you have the option, make the processes
as coarse as possible. Eg. I'm planning to run functions that take about
0.1 to 2 seconds to complete, so the microseconds spent communicating are
not relevent. I could also split up functions that take about 1/1000th of
that time, but then communication overhead becomes significant.

I'd probably go for the 50 single-processor machines, because writing the
software to use dual processors is more complex (or at least, seems more
complex - I've not tried anything serious yet). But that decision depends
on the problem domain.

One interesting thing I read - the more machines you have, the more often
you'll seem to have machine failures. So make sure the machines are easy to
replace, and that your application can adapt to machines coming and going.

Darren

---------------------------------------------------------------
Next Nomikai: 20 November, 19:30 Tengu TokyoEkiMae 03-3275-3691
Next Meeting: 12 December, 12:30 Tokyo Station Yaesu central gate
---------------------------------------------------------------
Sponsor: PHT, makers of TurboLinux http://www.pht.co.jp