Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!uflorida!gatech!hubcap!Eugene
From: eugene@eos.arc.nasa.gov (Eugene Miya)
Newsgroups: comp.parallel
Subject: Re: concurrent program visualization (sic)
Message-ID: <5373@hubcap.clemson.edu>
Date: 2 May 89 18:02:56 GMT
Sender: fpst@hubcap.clemson.edu
Lines: 15
Approved: parallel@hubcap.clemson.edu

I hope readers remember that more often then not, graphics terminals,
workstations on a net (thin or thick wire) become synchronization
bottlenecks when displaying cute little behavior displayers.
Unless systems are correctly designed, they become yet-another-Heisenflop
problem.  This unintended side-effect never appears directly in code,
is dependent on synchronous I/O, and is very insideous.  Getting around
cutesy displays is only one of the problems (particularly bad because
they look so good to funding sources).

Oh, the problem is obviously not as bad for smaller numbers of processors
than larger numbers if explicit tokens are passed (right?).  Also if
monitoring software can selectively drop data (another risk).

--eugene miya
  NASA Ames