Newsgroups: comp.parallel.mpi
From: Tom Skoglund <toms@tjeld.ii.uib.no>
Subject: Re: MPI_REQUEST_MAX
Organization: University of Bergen, Norway
Date: 29 May 1998 11:12:56 +0200
Message-ID: <yzebtshphif.fsf@tjeld.ii.uib.no>

salo@cray.com (Eric Salo) writes:

> Did either of you ever send us a bug report? Because I don't recall ever
> Also, which release(s) are you using? If you don't yet have it, the latest
> full release (MPI 3.1) is part of the MPT 1.2 product available at:
> 

I'll check what version we used at he time, and if we have a new one.

> (I would humbly offer the opinion that any application which needs to
> create more than 16384 simultaneous requests on any given MPI process
> is probably in need of a rewrite, although I would be very interested
> in hearing a strong counter-argument...)

In my case the requests were not simultaneous. I used
MPI_Recv_init(...), and then MPI_Start(..) and MPI_Wait(...) for a set
of servers and clients. The requests were freed when exiting. My
problems arose when I should do for example 10.000 runs for the same
problem. The number of simultaneous requests in my case should not be
to high.


> 
> If y'all could please post (or email) a simple test case which shows
> a resource leak in MPI, we'll get it fixed ASAP. And if you can do


I'll see if there still is a problem.


-- 

    ### Tom Skoglund * toms@ii.uib.no * http://www.ii.uib.no/~toms ###