Configuring Scientific Applications in a Heterogeneous
Distributed System
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
Scientific applications are often most naturally structured as a collection
of interacting software components executing on heterogeneous machines
connected by a network. For example, an application might consist of a
visualization tool on a graphics workstation that displays data generated
by a scientific simulation on a parallel machine, which in turn uses a
data repository on a storage server for input. Yet despite having the
structure of a distributed program, the support provided for configuring
applications in this domain has generally been minimal at best. Here, an
approach to configuring scientific applications based on the Schooner
interconnection system is described. One key aspect of Schooner is a
machine- and language-independent interface specification language that
is used to generate interface code to bind components into the application
and map them onto suitable host architectures. The other is a runtime
system that implements support for both static and dynamic configuration.
This paper describes the Schooner application model, outlines the method
of creating component interfaces, and describes the runtime system and its
various configuration options. In contrast with traditional configuration
management systems, Schooner is a communication system that provides
underlying support for the configuration requirements of scientific
applications rather than a general high-level tool for distributed
programming.
IEE Distributed Systems Engineering Journal 3,
3 (September 1996), 173-184.
compressed postscript (240K)
postscript (2.9M)
back to publications
page