Newsgroups: comp.parallel.pvm
From: krbailey@eos.ncsu.edu (Kendall Robert Bailey)
Subject: master/slave on same host
Organization: North Carolina State University
Date: 9 Feb 1995 15:58:46 GMT
Message-ID: <3hde3m$2je@taco.cc.ncsu.edu>



I'm running some pvm code in which it would be fairly efficient to 
have a slave task on the same machine as the master, as long as
I keep them from butting heads.  The slave tasks spend some time
idle waiting for 1 of 3 types of messages to arrive.  Since they
don't know which will arrive next, they cycle through three non-blocking
recieves until one shows up.  My question is, is this really inefficient?
It looks like when I've got both a master and one slave (of many) on
one machine, the slave gets as many clock cycles as the master while
its just waiting.  Would it be better to have a blocking receive in the
slave and pack an integer flag at the top of the message indicating which
type it is?  Would this free up more clock cycles for tasks doing
real work??

Thanks,
Kendall Bailey
email: krbailey@eos.ncsu.edu
North Carolina State University
Dept. of Applied Mathematics
-- 
   

