From: Richard Maine Newsgroups: comp.lang.fortran,comp.parallel.pvm Subject: Re: superlinear speedup, Date: 15 Dec 1998 09:54:00 -0800 Organization: NASA Dryden Message-Id: References: <3660D892.E1EDBF12@est.it> <755odb$fid@robin.mcnc.org> <7565mg$if5$1@netra.msu.montana.edu> Xref: ukc comp.lang.fortran:62198 comp.parallel.pvm:7877 I said: > I've seen superliner speedup in real life. Louis Glassy writes: > Do superlinear speedups happen for any problems that aren't > search problems? E.g., 2D FFT, solving a linear system of equations, etc? Yes. > A text I have [*] says.... More theory, it sounds like. I've not read any texts on the subject (its not really my area). I've just run applications. The applications I recall had no special features. Mostly dominated by solutions to differential equations. The applications weren't even multi-processor aware at all. The speedups I recall were just from running multiple instances of the application, each instance using a different data set. No, there was no communication between the applications to take advantage of common computations or any other form of "cheating". Possibly one could do better with multi-processor-aware applications, but mine weren't. I'm sorry, but nothing that anyone quotes out of any text is likely to convince me that behavior that I saw can't happen. (For that, you'd probably have to switch fields majorly and bring out a psychology text or some such). -- Richard Maine maine@altair.dfrc.nasa.gov