"Fehd, Ronald J. (PHPPO)" <rjf2@CDC.GOV> wrote [in part]: > the paper presented code which divided the tasks for each CPU > one such subset division was by use of mod operator > and another by expert use of an identifier: > CPU-1 gets all less than '33' > CPU-2 gets all equal to '33' > CPU-3 gets all greater than '33' > > interesting main point: > elapsed time may decrease > but CPU time may increase

Which is as we would normally expect from other multi-processing programs and processes. We split the tasks among several processors, so it takes less wallclock time. But we not only do all the same tasks as before, but now we also have to handle load-balancing (either in SAS or through an external OS process), task distribution to the CPUs, and re-integration of the outputs from the CPUs. So we typically see *total* CPU time go up. (Unless the OS was already doing load- balancing and multi-processing behind our backs.)

David -- David Cassell, CSC Cassell.David@epa.gov Senior computing specialist mathematical statistician