From: "Tony T. Warnock" <u091889@cic-mail.lanl.gov>
Newsgroups: comp.lang.fortran,comp.parallel.pvm
Subject: Re: superlinear speedup, (was: fortran77 programming)
Date: Mon, 07 Dec 1998 08:40:36 -0700
Organization: CIC-3
Message-Id: <366BF6F3.1E8DC934@cic-mail.lanl.gov>
References: <3660D892.E1EDBF12@est.it> <3665044D.A0F57E59@kings.uq.edu.au>
    <3665E491.7993355@flash.net> <y6k909mqww.fsf@tweedledumb.cygnus.com>
    <m21zmgnuau.fsf@blinky.bfr.co.il> <y6emqejtgq.fsf@tweedledumb.cygnus.com>
    <m2d85xi988.fsf@blinky.bfr.co.il> <y6yaokiw0r.fsf@tweedledumb.cygnus.com>
Reply-To: ttw@lanl.gov
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Xref: ukc comp.lang.fortran:61903 comp.parallel.pvm:7847


A simple argument (due to Vance Faber of LANL) shows that superlinear speedup is
not possible. Given an N times faster processor, one can have the processor
execute the first instruction of each of the N slower processors, then the second
instruction, etc. The faster processor will finish at least as fast as the slower
set of processors.

Most reported superlinear speedups are due to incorrect comparisons. For example,
N smaller processors may be given more cache than the faster processor so that
what looks like superlinear speedup is just due to a larger effective cache. Same
for I/O channels or total memory.

Tony