Newsgroups: comp.parallel.mpi,comp.parallel.pvm
From: lga@sma.ch (Laurent Gasser)
Subject: Re: MPI or PVM
Organization: Swiss Meteorological Institute
Date: 22 Oct 1996 07:19:22 GMT
Message-ID: <54hshqINN9di@maz4.sma.ch>

In article <54b6g8$4p1@sol.sun.csd.unb.ca>, k3is@fundy.unbsj.ca (DA) writes:
> Yan Xu (yanx@ifi.uio.no) wrote:
> 
> : If I have to choose between PVM and MPI system, which one should I
> : choose and why? I have heared that the MPI system is better (quicker)
> : than PVM, is that so? Where can I find some references on this?
> 
> PVM has one implementation.....yes I found it is slower than the implementation
> of MPI that I used to solve one parallel problem.

PVM also has more than one implementation.  At least on CRAY, you get three
flavours (the public version, a version tune to T3D, a stand-alone version
using shared memory copying instead of deamons).  

You can expect to see very different performances from each version.  This is 
similar to MPI people saying that you should require from your hardware vendor
to produce a finely tuned implementation.

I was told that PVM people participated to the early definition of MPI.  They
had aknowledge deficiences in their library and helped at defining the MPI
standard.  Where could I get the rational for weakness of PVM justifying
the definition of MPI?

-- 
Laurent Gasser (lga@sma.ch)
Computers do not solve problems, they execute solutions.



