Affordable High Performance Computing Cooperative Agreement


Objective: This program will demonstrate networked workstation performance
and reliability at levels equivalent to 1994 Vector Supercomputers at 1/4 of
the capital cost.  This program will also support two additional CAS Level 1
milestones: demonstrate end-to-end reductions in cost and time to solution
for aerospace design applications on heterogeneous systems; and demonstrate
portable, scalable programming and runtime environment for Grand Challenge
Applications on TeraFLOPS scalable system.

Approach: Two aerospace propulsion application problems will be addressed: a
three-dimensional steady flow Navier-Stokes simulation of a complete
high-pressure compressor, and a procedure for predicting resonant stress
attributable to rotor/stator interaction. Overnight turnaround for the two
application problems will enable the design of advanced engines at lower
cost and with reduced time to market. To achieve this overnight turnaround,
a task by task improvement of propulsion system applications, system
software, and hardware has been identified. Visualization, message passing,
and system software developed under this project will be tested for
reliability, efficiency, and ease of use. A commercialization plan will be
prepared to allow for broad-based industry use of selected software.
Portability and performance of the software developed will be ensured by
collaboration with several major UNIX workstation vendors. A team of
representatives from the propulsion industry, universities, NASA centers,
and software and hardware vendors will carry out the tasks.

Accomplishment: Speedup of the CFD solver code for multistage simulation
has reached 2X. Interactive viewing of distributed applications using the
pV3-Gold visualization package is available in a beta-test version. Examples
demonstrating the package featuring two MPEG movies are located on the
World Wide Web at URL: http://danville.res.utc.com/AHPCP/movie.htm . Scheduling
of parallel jobs that span multiple clusters has been demonstrated.

Significance: Achievement of overnight turnaround for complex aerospace
propulsion applications will lead to reduced product cost and shorter time
to market. In addition, commercially available system software will
potentially benefit a broad range of industries from financial management to
oil exploration.

Status/Plans: All elements of the AHPC project are on schedule. Job migration
and dynamic parallel job scheduling will be implemented by 12/31/96. The High
Performance Distributed File System will be implemented by 3/31/97. Code
coupling of the CFD applications used in this project will be demonstrated
with the Visual Computing Environment (VCE) by 6/30/97. Multicluster job
scheduling with load and capacity parameters across Wide Area Networks (WANs)
will be demonstrated by 6/30/97. A full system demonstration of a one day
turnaround for a multistage compressor analysis on an affordable computing
testbed will be demonstrated by 9/30/97.

Point(s) of Contact:

Theresa Babrauckas
NASA Lewis Research Center
21000 Brookpark Rd. MS 142-4
Cleveland, OH 44135
theresa@lerc.nasa.gov
(216)433-8723

[Image]
pV3-Gold screen: /usr/lerc/fstbab/ito/rtop/pv3.tif

Also have MPEG movies: /usr/lerc/fstbab/ito/rtop/pv3a.mpeg and
                       /usr/lerc/fstbab/ito/rtop/pv3b.mpeg