Newsgroups: comp.parallel.mpi
From: salo@mrjones.engr.sgi.com (Eric Salo)
Subject: Re: Behaviour of MPI/P4 on uniprocessors....
Organization: Silicon Graphics, Inc.  Mountain View, CA
Date: 12 Mar 1997 00:30:35 GMT
Message-ID: <5g4tfb$2sp@murrow.corp.sgi.com>

My guess is that this is simply due to the polling nature of the p4
shmem device. You've only got one CPU, right? So even though you're
running 2, 4, 8 or 16 MPI processes, ultimately they must all contend
for that same CPU. But in many shared memory implementations of MPI,
processes do not put themselves to sleep when they have nothing to do;
they just spin on a shared memory queue/semaphore, so as far as the OS
is concerned they are still doing real work and will therefore not be
descheduled early.

Eric Salo    Silicon Graphics Inc.    salo@sgi.com