From plassman@cse.psu.edu Sun Jun  2 15:53:16 2002
Date: Thu, 16 May 2002 11:30:21 -0400 (EDT)
From: Paul E Plassmann <plassman@cse.psu.edu>
To: plassman@cse.psu.edu, lnl@psu.edu
Cc: nxs216@psu.edu, anirudh@anirudh.net
Subject: Re: new version of paper

Some side notes for future work:

Yes, the simulations are 4D, but they are really 3D X 1D
where the time dimension is currently only being sampled
at the time step being computed.  One thing that we haven't
taken advantage of is sampling in time along a lower dimensional
space.  That is, we could pick an 2D surface or a 1D slice
and time-average (or any other time-dependent metric) on
that lower-dimensional surface.  This is also "scalable"
and uses the dimensional reduction in a different manner.
I imagine that this sort of information would be of great
interest for unsteady flows.

Another way that the time-dependent data could be used is 
to archive data on the local disk.  Then it would be possible
to do time-dependent queries from stored time slices in
a more scalable way.  Of course, we would have to deal
with the overhead of disk reads/writes, but we can deal with
time histories this way.

You could also use the archives for efficient restart to
get data from any intermediate time.  Rather than having
to completely restart the simulation, you could jump around
in time with much less overhead than you would have if you
had to start the simulation from scratch.  This would allow
a whole new "dimension" to the monitoring.

Paul
