A Software Platform for Constructing Scientific Applications
from Heterogeneous Resources
Patrick T. Homer
and
Richard D. Schlichting
Department of Computer Science
The University of Arizona
Tucson, Arizona 85721, USA
{patrick, rick}@cs.arizona.edu
Abstract
Support for heterogeneous processing is useful for
increasing the functionality available to designers of
scientific applications.
For example, rather than implement an application requiring remote
vector processing and local visualization as two separate programs,
such support allows an alternative structure in which the application
is a single logical program with transparent transfer of control and
data between phases.
In addition to being simpler and more intuitive, such
structuring makes it feasible to enhance the way in which users
interact with the application to do, for instance, model steering.
Here, a software platform that facilitates the construction of
this type of scientific application is described.
Its key component is Schooner, an interconnection system
that includes an intermediate data representation, a simple
specification language, and a heterogeneous remote procedure call (RPC)
facility; to provide sophisticated visualization capabilities and an execution
framework, AVS is included as well.
Two applications built using this platform, one from molecular dynamics
and the other involving neural nets, are also described.
One important conclusion is that enhanced monitoring and interaction
facilities impose very little overhead for applications such as these.
Journal of Parallel and Distributed Computing 21, 3,
(June 1994), 301-315.
compressed postscript (120K)
postscript (330K)
back to publications
page