Newsgroups: comp.parallel.mpi From: 110022.718@compuserve.com (Shane Hebert) Subject: Re: MPI doesn't perform well on the Cray J90 Organization: Mississippi State University Date: Fri, 26 Sep 1997 19:36:30 GMT Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Message-ID: <342c0db6.356582328@nntp.msstate.edu> On Thu, 25 Sep 1997 17:23:35 -0700, an alien in the guise of Roberto Lionello wrote: >| Hello, >| >| I'm doing some tests on a J90 using a program developed on the >| T3E. With a given size of the problem, it scales nicely on the T3E >| using up to 128 processors. I compiled it on the J90 using >| TCP sockets and I was disappointed by the performance: >| it saturates with 8 processors! :-( >| Why the performance of MPI is so poor on the J90? Is it because >| they didn't bother to write a good implementation of MPI? >| I thought that having the memory in a single block >| instead of being divided between the nodes would have been an >| advantage (forgive me if I'm being naive) >| When I specify "mpirun -np 8 myprog" do I really get 8 processors >| only for myself? >| Thanks! One of the reasons it is slower is that you are using TCP communications. I don't know anything about the native Cray implementation of MPI for the J90 or about any native J90 device code for MPICH (or even if either exists). I would suggest that you try to find a machine specific version of MPI or J90 device code for MPICH and run your program again. Your performance should get better. Shane ============================================================ Shane Hebert shane@erc.msstate.edu You must use this email address to contact me as the address in my reply-to is that of the last unsolicited email that I received (this time a group selling email addresses). Disclaimer: Any opinions stated above are mine and not those of my employer.