Path: utzoo!utgpu!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!unmvax!gatech!hubcap!anand
From: anand@amax.npac.syr.edu (Anand Rangachari)
Newsgroups: comp.parallel
Subject: Superlinear speedup
Message-ID: <3842@hubcap.UUCP>
Date: 12 Dec 88 13:18:39 GMT
Sender: fpst@hubcap.UUCP
Lines: 22
Approved: parallel@hubcap.clemson.edu


  This may seem obvious but in order to measure the speedup of a program
it is neccessay to be able to run the given program with differing
numbers of processors.

  I have seen some programs developed on the Encore Multimax here and 
have been apalled to find that the number of processors used for some 
programs is hardwired in. ie nothing less than major surgery would be 
needed to do a proper speedup measurement.

  Also, I note that when the Sandia group achieved their prodigious
speedups, they were unable to measure the single processor case. Two 
reasons were mentioned for this. Firstly with a speedup of say 500
on 1000 processors a program taking 10 min to run will need 5000 min
on a single processor. Secondly it may not be possible to fit the 
entire problem on a single processor in the case of distributed
machines with a limited memory for each processor.

                                                   R. Anand

Internet: anand@amax.npac.syr.edu
Bitnet:   ranand@sunrise