To: Dr. R. E. Reynolds
From: Mitch Rosen
Cc: Haydn Wadley
Date: July 5, 2000
Re: HIGH PERFORMANCE COMPUTING AT THE UNIVERSITY OF VIRGINIA:
A RECOMMENDATION
AS A KEY
INFORMATION TECHNOLOGY THRUST AREA FOR 2000 - 2001
The University has centrally supported high performance computing
(HPC) systems for shared use for many years, and they have been in
great demand. Given the directions in molecular & genetic modeling,
for example, and other computationally intensive work of our top
researchers, additional investments in leading edge technologies are
necessary.
Where is the University headed?
Where should the University put its resources?
The 2020 Science & Technology Planning Commission recognized, too,
that very significant infrastructure expenditures are required. but
the time frame for 2020 implementation is unknown.
The HPC arena is changing, and needs to be addressed in light of
current and projected researchers' needs.
The following points are of concern:
-The demands for high end computing continue to grow.
-Traditional vector and array machines continue to be marketed, but
there are emerging, new technologies such as the Linux cluster
supercomputer (i.e., Beowulf) and Prof. Grimshaw's [CS Dept.]
Legion software & Centurion hardware that may have appeal to some
researchers.
-There is a changing environment at the low end. The competitive
performance of standalone workstations and PCs has improved. That
combined with slow overall turnaround times on the shared
University systems has lead some researchers to work in a local,
distributed mode.
-The flagship of the University's computational horsepower is the
IBM SP2 system. However, with IBM announcing the end of the
UVA-IBM Alliance / Shared University Research (SUR) program, which
has allowed regular system upgrades at special rates, the future of
this machine needs to be determined. Under the SUR program of Fall
1998, the SP2 computer was last upgraded two years ago with
additional memory and some new CPUs.
Perhaps a University Strategic Plan for HPC is needed.
-How about preparing a "Guidebook" for University researchers with
the goal of outlining what platforms and software, and computational
performance, will be available on a shared basis--- so they can
coordinate their local investments and select compatible software and
hardware acquired from grant and departmental funds.
Who should examine this issue?
-What systems and support are required to meet researchers'
needs?
-Best bang for the buck?
-Centralized or distributed?
-Cost sharing (University, federal, industrial, etc.)?
How to get the discussion going?. Who should be included?
-The main user base from Engineering, Arts & Sciences and
Medicine, and of course, ITC, and the oversight Committees (UCIT,
ACAC, etc.).
Sample list of heavy HPC users are:
ENGINEERING:
Materials Science, Mechanical Engineering, Biomedical
Engineering, Civil Engineering
ARTS & SCIENCES:
Astronomy, Economics, Chemistry,
MEDICINE:
Biochemistry
I look forward to following up with you on these thoughts.