Newsgroups: comp.parallel.pvm,comp.parallel.mpi
From: csundqvi@abo.fi (Christoffer Sundqvist)
Subject: Re: PVM send buffers
Organization: Department of Business Administration
Date: Wed, 20 Mar 96 11:36:57 GMT
Message-ID: <4iog5n$n20@josie.abo.fi>

In article <4ik7rf$g9k@newshost.lanl.gov>,
   rbarrett@ranger.lanl.gov (Richard Barrett) wrote:
>Folks,
>
>A week or so ago I posted regarding the pvm_pack functions. I asked if
>others would like to be able to allocate the entire send buffer before
>making multiple calls to pvm_pack in order to reduce the overhead of
>malloc. (This would be analogous to MPI_Buffer_attach.)
>
>So far I count zero responses, and I didn't receive any via private email.
>
Hi,
I tried to mail you but it bounced (user unknown).
We are working on a parallel interior point method for linear optimization and 
we need to call pvm_pack repeteadly in several places in our code. So what you 
are talking about sounds very intresting. Our actual problem is that we need 
to partition a large matrix in smaller parts to send it over the mesh. We are 
using C and dynamically allocated arrays. I posted a question about this some 
weeks ago but didn't receive any answer either. Here the problem is that 
pvm_pack in C on our Convex could only pack a dynamic array if it is 
in a continous block of memory, which isn't the case when we want a 
subblock from the middle of a larger matrix. Instead we are packing one
row of the matrix at a time so if there is any way of reducing the overhead
of calling pvm_pack repeatedly it could greatly speed up our code.

Christoffer Sundqvist

email csundqvi@abo.fi
http://www.abo.fi/~csundqvi


