From: hjstein@bfr.co.il (Harvey J. Stein) Newsgroups: comp.lang.fortran,comp.parallel.pvm Subject: Re: fortran77 programming Date: 07 Dec 1998 11:01:32 +0200 Organization: Unspecified Organization Message-Id: References: <3660D892.E1EDBF12@est.it> <3665044D.A0F57E59@kings.uq.edu.au> <3665E491.7993355@flash.net> Cc: hjstein@bfr.co.il Xref: ukc comp.lang.fortran:61896 comp.parallel.pvm:7845 Craig Burley writes: > hjstein@bfr.co.il (Harvey J. Stein) writes: > > Doesn't matter. You won't get superlinear speedups by using parallel > hardware. I don't care whether your application is real-time or not. I came into the middle of this when I saw your post on comp.parallel.pvm. It seems I might have missed the point that it's a long discussion about superlinear speedups in multi-cpu boxes. In any case, I can't really abide by your "all else being equal" assumptions, because when talking about multi-cpu SMP boxes, one typically means N CPUs in 1 box with appropriate support hardware vs 1 of the same CPUs in 1 box with appropriate support hardware. One also typically requires the same OS, etc on each system. This rules out things like letting the single cpu system have the same amt of total cache & same total # of registers as the N CPU box. Under these conditions I just wanted to point out that one could imagine the N CPU box giving better performance than the single CPU box (containing a CPU running N times faster) because there could be fewer context switches with the single CPU box. Of course, I'd expect that the difference in performance would be close to unmeasurable because I'd expect the context switch overhead to be extremely small, but it should still be there. -- Harvey J. Stein BFM Financial Research hjstein@bfr.co.il