Newsgroups: comp.parallel
From: shi@astro.ocis.temple.edu (Yuan Shi)
Subject: Re: superlinear speedup ref?
Organization: Temple University, Academic Computer Services
Date: Sat, 5 Aug 1995 14:01:13 GMT
Message-ID: <3vvtj9$a9r@cronkite.ocis.temple.edu>

G. Evans (gme@sys.uea.ac.uk) wrote:
: In article <Ted.Belding-1207951713260001@pm054-25.dialip.mich.net>, Ted.Belding@umich.edu (Theodore C. Belding) writes:
: |> I'm looking for a reference concerning superlinear speedup 
: |> in search algorithms.  Specifically: the type of superlinear speedup 
: |> obtained by running n identical serial search programs concurrently 
: |> on the same problem.  (The wall clock time to solution decreases faster
: |> than 1/n, under certain conditions).  Any pointers would be much 
: |> appreciated.  TIA!
: |> -Ted
: |> 
: |> --

: Search anomalies (including superlinear speedup) in parallel Branch-and-Bound
: algorithms are described in, for example,

: T. Lai and S. Sahni, Anomalies in Parallel Branch-and-Bound Algorithms,
: Communications of the ACM, 27, 1984.

: I've got a number of papers on parallel Branch-and-Bound which actually
: report superlinear. I can give you references in you want them.

: Gareth.

Speedup is a relative measure that depends largely on your prefered definitions.You can get superlinear speedup for almost any algorithm if you include 
resource related factors in your speedup base definition. There are also
serial algorithms that will give superlinear speedups even if you exclude
resource factors, namly their parallel implementation(s) will require less
number of calculation steps as compared to the original serial algorithm.

I have a paper that explains these problems in detail. You are welcome to a 
copy via http://argo.cis.temple.edu.(under the name "Timing models ...").

Yuan

shi@falcon.cis.temple.edu


