Newsgroups: comp.parallel.pvm
From: huebner@convex.com (Robert Huebner)
Subject: Re: problems with pvm_reduce
Organization: CONVEX News Network, Engineering (cnn.eng), Richardson, Tx USA
Date: 14 Jul 94 15:10:40 GMT
Message-ID: <huebner.774198640@cnn.eng.convex.com>

In <302teo$cvd@brachio.zrz.TU-Berlin.DE> iste1030@w250zrz.zrz.tu-berlin.de (Ivan Santibanez) writes:

>	Using pvm.3.3.2 I got some problems using pvm_reduce: It was not
>possible to made the correct global sum of all members. I  think this a
>bug because  also the examples in the distribution don`t work. Are
>there some other experiences, or could somebody advice me how
>can I bring pvm_reduce to work correct ?

I have also seen strange behavior with pvm_reduce.  Basically it seems
that the result code goes to some random (well, not totally random but
dependant on the number of participants) task rather than the test specified
in the 'root' parameter.  If you print out the resulting arrays from all
participants, you'll find someone else probably has the results.  Occasionally
no one seems to have the result, though.  With two processes it seems
no results are delivered.

