From: H Richardson <harvey@tromba.demon.co.uk>
Newsgroups: comp.parallel.mpi
Subject: Re: MPI Profiling Interface hassle
Date: Wed, 16 Jun 1999 09:45:43 +0100
Message-Id: <37676437.B346FDF3@tromba.demon.co.uk>
References: <37651080.BE3BD2EC@uk.sun.com> <7k3opj$2bj6@news2.newsguy.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Xref: ukc comp.parallel.mpi:5211


Greg Lindahl wrote:
> 
> If you keep track on the send side, you will have an accurate
> count. You also know the destination of all sends, so you can count
> the bytes received.
> 

Yes I can and do track the sends, and optionally the destinations.
This does not help on the receive.  The send could be matched
by a blocking or nonblocking receive.  Blocking is fine.
Nonblocking means that I have to attribute the communication to 
the Wait/Test call that completes the receive.  But for reasons
previously described that is what is hard to do.

Harvey