Newsgroups: comp.parallel
From: montoya@mrco.carleton.ca (Frederic Montoya)
Subject: Re: Help - explain superlinear speedup?
Organization: Carleton University
Date: Mon, 16 Jan 1995 17:23:37 GMT
Message-ID: <D2I83H.Dw2@cunews.carleton.ca>


In article <3fde3h$3cm@infosrv.edvz.univie.ac.at> you wrote:
> In article <3f23ht$ec9@agate.berkeley.edu>, slater@nuc.berkeley.edu (Steve Slater) writes:
> |> 
> |> I have a program which has superlinear speedup and I
> |> can't explain it. Does anyone have any ideas. Here is
> |> the summary.

> I think what really counts is comparing the multiprocessor version with
> a single processor, single process version executed on a single SPARC, right?

It would even be better to compare with a truely sequential code 
(I mean, optimized for sequential execution and not only the single process 
version of the parallel code).
It's quite clear that in this case, cache effects (and TLB probably) and
context switches are the only reasonable way to account for the super-lienar 
speed-up. 

Frederic.




