Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!uunet!ncrlnk!ncrcae!hubcap!bjornl
From: bjornl@tds.kth.se (Bj|rn Lisper)
Newsgroups: comp.parallel
Subject: Re: IPSC Communications
Message-ID: <7203@hubcap.clemson.edu>
Date: 27 Nov 89 13:40:24 GMT
Sender: fpst@hubcap.clemson.edu
Lines: 24
Approved: parallel@hubcap.clemson.edu

In article <7161@hubcap.clemson.edu> parker@vienna3.tmc.edu (Bruce Parker)
writes:
%In article <7143@hubcap.clemson.edu> argosy!workman@decwrl.dec.com (Will
Workman) writes:
%>The CM-2 really suffers a performance penalty by not having
%>a physical grid interconnect system on applications like
%>image and signal processing which are heavily dependent on
%>fast nearest neighbor and reduction techniques.

%Those of you who are less biased towards selling a
%particular architecture might consider that these guys may
%not have the best mapping of mesh nodes to hypercube nodes.
%There does exist a "real-time" (O(1) delay) emulation of a
%mesh on a butterfly network ....  Hypercube emulation
%of a butterfly is trivial (simulate each column by a single
%node), although the load is now O(lg n). .... but it may be possible to
%sidestep the butterfly-hypercube emulation and implement the
%emulation directly on the hypercube.

I was under the impression that the CM-2 uses binary-reflected Gray encoding
of the grid coordinates. That gives a direct embedding of the grid into the
hypercube topology. 

Bjorn Lisper


Brought to you by Super Global Mega Corp .com