[Beowulf] cluster softwares supporting parallel CFD computing

Eric W. Biederman ebiederm at xmission.com
Fri Sep 8 20:58:58 PDT 2006


"Vincent Diepeveen" <diep at xs4all.nl> writes:

> You're assuming that you run 1 thread at a 2 to 4 core node or so?

Not at all.  I am assuming 1 thread per core, is typical.

So if you have 4 cores and 4 threads, when one of them is asleep.

Most likely when the interrupt arrives your previous cpu
will be free so you will wake up immediately, and start executing code
again.  

If there is not a cpu free then some daemon is running or some
operating system task is running and doing useful work.   So
you have a win because it ran when you didn't need the cpu.
Most likely whatever it is will not be cpu intensive and will
relinquish the cpu before it's time slice and your mpi
job will run again.  In the worst case you have to wait an
entire scheduler time slice.

Plus it doesn't really matter because before you fall asleep your
job has been waiting for a relatively long time.  So getting
moving a few extra microseconds later is in the noise.

Eric



More information about the Beowulf mailing list