From plassman@cse.psu.edu Tue May 21 18:43:27 2002
Date: Thu, 16 May 2002 11:03:36 -0400 (EDT)
From: Paul E Plassmann <plassman@cse.psu.edu>
To: nxs216@psu.edu, lnl@psu.edu
Cc: anirudh@anirudh.net, plassman@cse.psu.edu
Subject: Re: new version of paper

Some notes:

1.  We say (in the title) that this is a "scalable" computational
steering system, but we never say why.   The reason is that
the data required for the CFD simulation lives in higher
dimensional space (3D) than the data that we visualize
(isosurfaces (2D) and chord plots (1D)).  The scalability
comes from this "dimensional reduction" that is done on
parallel machine (in a scalable manner) rather the traditional
(non-scalable) way of post-processing a file.

1b.  We should have some numbers comparing the size of a
file containing the simulation data vs. the size range of
data for the iso-surfaces that we request.

2.  We might also make the case that this approach is
scalable both in "space" and "time".  That is, by monitoring
a time-dependent simulation we have access to the whole
time history, whereas it would be prohibitive to store the
time history in a bunch of files.  We could even say how
much space this would take up.

3.  We introduce the COCOA clusters twice, on the top of
page 2 and the middle right of page 7.

4.  At the bottom right of page 2 we introduce the computational
steering phases:  instrumentation, monitoring, + steering.
We should include a reference here.

5.  We use the term "lightweight" but we don't make it clear
what this means.  The idea is that the computational overhead
of the server thread is very small (in terms of memory + CPU).
Can we say how small when there are no queries?   Obviously, if
there are requests, there will be overhead.  Can we say something
about the cost of computing an isosurface?  What about running
on a two-processor node, when can the two threads use an SMP
to minimize the overhead of the steering system.

6.  Figure 2.  The "server" code and the "client" code need
to be in separate figures.  Otherwise, people will think this
is in the same program.

7.  In the iso-surface routine section, we should say that the
iso-surface algorithm is basically the marching cubes alg. applied
to tetrahedra.  I don't know if there is a reference for this.

Paul
