Newsgroups: comp.parallel From: christen@mint.ifi.unibas.ch (Peter Christen) Subject: Timing on SP-2 Organization: Date: 11 Jun 1998 11:01:26 GMT Message-ID: <6lode6$9r3$1@encore.ece.cmu.edu> Hi there, I did some timing measurements on an IBM SP-2 (16 nodes) with MPI and got really strange results. I measured a computational block in my program (that performed many floating-point operations), both with MPI_Wtime() and gettimeofday(). On one node, the time returned by MPI_Wtime() was exactly the time returned by gettimeofday() divided by 50. On another node, the MPI_Wtime() was the same as the gettimeofday() divided by 500! Why this? Does anybody know? Secondly, I wanted to know how other jobs running at the same time on the same node affect the runtime of my program; so I started other floating-point intensive jobs (up to 5 at the same time). The times I measured in my program with MPI_Wtime() and gettimeofday() are almost the same as when no other job is running! Running five computational intensive jobs and my program at the same time only yields in an increased measured runtime of my program of up to 20%. On a Sparc workstation, the measured runtime is roughly doubled if another job is running at the same time (this is what I expect!). As far as I know, gettimeofday() returns the system time, and I expected a much larger increase in runtime if so many other jobs are running concurrently. Has anybody received similar results? Can anybody explain this strange results? Thanks, Peter -------------- Peter Christen Dipl. Informatik-Ing. ETH Institut fuer Informatik tel: +41 (0)61 321 99 67 Universitaet Basel fax: +41 (0)61 321 99 15 Mittlere Strasse 142 CH-4056 Basel email: christen@ifi.unibas.ch Switzerland www: http://www.ifi.unibas.ch/~christen -- Articles to bigrigg+parallel@cs.cmu.edu (Admin: bigrigg@cs.cmu.edu) Archive: http://www.hensa.ac.uk/parallel/internet/usenet/comp.parallel