From Interactive High Performance Programs to Distributed Laboratories:
A Research Agenda
Greg Eisenhauer, Beth Schroeder and Karsten Schwan
Abstract
Many scientific problems require highly complex and difficult computations.
If the individuals who study these problems can interact with the
computational tools while the computations are progressing, the analysis of
the problem could be much more efficient and effective. This article
introduces the potential increases in functionality and performance gained by
the online interaction of end users with high performance computational
instruments on single and on networked parallel machines (also
see
[sch96],
). We consider systems in which users interact with
computations as if they were physically accessible laboratory instruments and
in which entire distributed laboratories are constructed from sets of
such computational instruments. Within this context, our intent is to
facilitate both online interactions with single computational instruments and
interactions among multiple scientists and multiple instruments located at
physically distributed sites where scientists may have dissimilar areas of
expertise.
With our research and with the larger-scale Distributed Laboratories
project at the Georgia Institute of Technology, we aim to
improve the state of the art of interactive high performance computing for
parallel and distributed applications on the variety of heterogeneous
platforms now in common use by HPC users and researchers. Particularly, our
goal is to develop a general framework for enhancing the interactivity of high
performance applications. This framework:
-
contains general interactivity (i.e., online monitoring and steering)
mechanisms with which high performance applications may be inspected and/or
steered at runtime by algorithms, human users, or both;
-
supports interchangeable visualizations run across heterogeneous
and distributed hardware platforms, using a robust and portable
data meta-format for transporting visualization content;
-
supports the simultaneous interaction of multiple scientists with single
large-scale computations; and
-
permits collaboration via multiple computational instruments among scientists
working in separate locations.
In this article we introduce one of the
parallel and distributed scientific applications used in our Distributed
Laboratory
research. We then explore interactions with a single
computational instrument during its execution,
typically referred to as interactive program
steering. Finally we
discuss the use of multiple computational
instruments by sets of end users, thereby moving from issues
addressed in previous work toward interesting topics for future
research.
Availability
This paper is appear in the Proceedings of the SPDP'96 Workshop on Program
Visualization and Instrumentation and is available prior to publication in
PostScript
form,
Bibtex Entry
@InProceedings{Eisenhauer96FIH,
author = "Greg Eisenhauer and Beth Schroeder
and Karsten Schwan",
title = "From Interactive High Performance Programs
to Distributed Laboratories: A Research Agenda",
booktitle = "Proceedings of the SPDP'96 Workshop on Program
Visualization and Instrumentation",
year = "1996",
month = "October",
}
Related Links
The home pages for the
Falcon project
and the
Distributed Laboratories project may also
be of interest.
Maintained by:
Greg Eisenhauer
College of Computing
Georgia Institute of Technology
eisen.nospam@cc.gatech.edu