Path: utzoo!mnetor!uunet!husc6!bloom-beacon!gatech!hubcap!uiucdcs!pur-ee!uiucdcsm.cs.uiuc.edu!grunwald
From: uiucdcs!pur-ee!uiucdcsm.cs.uiuc.edu!grunwald@uunet.uu.net
Newsgroups: comp.parallel
Subject: Re: Sandia scaling results
Message-ID: <1254@hubcap.UUCP>
Date: 30 Mar 88 13:23:16 GMT
Sender: fpst@hubcap.UUCP
Lines: 16
Approved: parallel@hubcap.clemson.edu

I've read the paper & basically, there's nothing that wonderful about what they
did (no slight intended to the folks at Sandia).

They took a parallel algorithm, did a good implementation, measured
the results and analysed them.

The reason they're getting the press instead of similar efforts at Rutgers, etc
is that they had a 1024 processor system. This allows them to show real speedup
of 1020, where you can do 128 at best.

Now, you might say ``So? They've got money and we don't'' -- and you'd be
right. However, this was the requirement of the Karp Challange. Real Speedup
on Real Hardware.

So basically, they've shown that Amdahls law, while true, isn't so limiting
for certain problems, and that speedup in the 1000's is possible.