Path: utzoo!utgpu!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!rutgers!gatech!hubcap!eos!eugene
From: eos!eugene@eos.arc.nasa.gov (Eugene Miya)
Newsgroups: comp.parallel
Subject: Re: Superlinear speedup
Message-ID: <3847@hubcap.UUCP>
Date: 12 Dec 88 18:02:52 GMT
Sender: fpst@hubcap.UUCP
Lines: 33
Approved: parallel@hubcap.clemson.edu

In article <3841@hubcap.UUCP> gusciora@sam.cs.cmu.edu (George Gusciora) writes:
>Why is everyone so surprised about superlinear speedup?

The above coming from a person at the University where it was originally
written up.

"Speed up," let's face it, is a dumb measure.  Users, given less than
half a chance would not care about the n-1 cases leading up to maximum
performance, they only care about the nth, "highest performance case."
Only computer scientists care immediately about the progression.
A couple of years ago, I can't remember the paper, someone wrote that
computists don't induct: we don't completely study all there is know about
2-processors before moving to 3-processors, etc.  We don't learn a whole
which is transferable between cases.

George also noted the WARP. The best I've heard for a 10 processor system
is 85 MFLOPS on a matrix multiply (at CMU).

The problem isn't parallelism, it's the sequential stuff (within and
without the "parallel streams.")

Another gross generalization from

--eugene miya, NASA Ames Research Center, eugene@aurora.arc.nasa.gov
  resident cynic at the Rock of Ages Home for Retired Hackers:
  "Mailers?! HA!", "If my mail does not reach you, please accept my apology."
  {uunet,hplabs,ncar,decwrl,allegra,tektronix}!ames!aurora!eugene
  "Send mail, avoid follow-ups.  If enough, I'll summarize."

  If you want a cute little exercise work a week without using the
phrase "parallel processing."  Then excise the other common terms:
massive parallelism, multiprocess, multitask, etc.  Have fun.
You will learn something.