Path: utzoo!news-server.csri.toronto.edu!cs.utexas.edu!uunet!lll-winken!llnl!physics.llnl.gov!brooks
From: brooks@physics.llnl.gov (Eugene D. Brooks III)
Newsgroups: comp.arch
Subject: Re: massive parallelism, was  CDC 6600 and TI ASC
Message-ID: <781@llnl.LLNL.GOV>
Date: 11 Mar 91 02:06:49 GMT
References: <1991Mar7.215545.430@zoo.toronto.edu> <LAMSON.91Mar7181743@el1.crd.ge.com> <13834@life.ai.mit.edu>
Sender: news@llnl.LLNL.GOV
Organization: Lawrence Livermore National Laboratory
Lines: 10

In article <13834@life.ai.mit.edu> lethin@raisin-scone.ai.mit.edu (Richard A. Lethin) writes:
>the CM has more wires: more communication.  While communication steps
>can be very costly, they don't need to pass through the bottleneck of
>a Central processing unit as they might when doing a scatter/gather on
>a vector machine.
Scatter/gather is the equivalent of the communication on the CM.  On
a Cray XMP or YMP the scatter/gather bandwidth matches the cpu speed
just fine and does not represent a bottleneck.  It is certainly not
as hampered as the "router" functionality on the CM-2 which is a severe
mismatch for the PE computational bandwidth.