Newsgroups: comp.parallel.mpi
From: vgeor@leo.unm.edu (Vasilis Georgitsis)
Subject: Question about implementation of Collective functions.
Keywords: Collective
Organization: University of New Mexico, Albuquerque
Date: 13 Jun 1995 17:46:31 GMT
Message-ID: <3rkitn$qs9@lynx.unm.edu>

Hi there,
  I am a student at UNM and I am working with a number of message passing
libraries that are running on an SP2 machine. I am looking for expressions
that would give the programmer the ability to predict the communication time,
and the bandwidth of a collective function that he or she would used on his
or her application.  Does the MPIF or any other message passing library has
any formulas that show the expected performance of each of the collective
function the library provides? I have tried to look into the source code of
MPICH but I could not be able to come up with any expression.  Zhiwei Xu
posted an e-mail on 21st of April, 1995 with the title "Is SP2 the best
supercomputer? (Long)".  He provided some expression for a number of function
of MPL Operations. On the second section of that mail he mentioned that the
vendors could done better and more accurate because they are in the best
position to do a controlled experiments and at the same time they know the
form of each of the expression.  Is/are any vendors that have that avaliable
and if so could someone point me towards that direction?

Thanks in advance
Vasilios Georgitsis

