Modelling Parallel and Distributed Virtual Reality Systems for Performance Analysis and Comparison
Abstract
Most Virtual Reality systems employ some form of parallel processing, making
use of multiple processors which are often distributed over large areas geographically,
and which communicate via various forms of message passing. The approaches to
parallel decomposition differ for each system, as do the performance implications
of each approach. Previous comparisons have only identified and categorized
the different approaches. None have examined the performance issues involved
in the different parallel decompositions. Performance measurement for a Virtual
Reality system differs from that of other parallel systems in that some measure
of the delays involved with the interaction of the separate components is required,
in addition to the measure of the throughput of the system. Existing performance
analysis approaches are typically not well suited to providing both these measures.
This thesis describes the development of a performance analysis technique that
is able to provide measures of both interaction latency and cycle time for a
model of a Virtual Reality system. This technique allows performance measures
to be generated as symbolic expressions describing the relationships between
the delays in the model. It automatically generates constraint regions, specifying
the values of the system parameters for which performance characteristics change.
The performance analysis technique shows strong agreement with values measured
from implementation of three common decomposition strategies on two message
passing architectures.
The technique is successfully applied to a range of parallel decomposition strategies
found in Parallel and Distributed Virtual Reality systems. For each system,
the primary decomposition techniques are isolated and analysed to determine
their performance characteristics. This analysis allows a comparison of the
various decomposition techniques, and in many cases reveals trends in their
behaviour that would have gone unnoticed with alternative analysis techniques.
The work described in this thesis supports the Performance Analysis and Comparison
of Parallel and Distributed Virtual Reality systems. In addition it acts as
a reference, describing the performance characteristics of decomposition strategies
used in Virtual Reality systems.
Participants
-
Shaun Bangay
-
Peter Clayton
-
David Sewry
Conference Papers (peer reviewed)
[1] Shaun Bangay, Peter Clayton, and David Sewry. Analytical simulation for performance analysis of distributed virtual reality systems. In SS '97: Proceedings of the 30th Annual Simulation Symposium (SS '97), pages 206–215, April 1997. [PDF] [BibTeX]
PhD Theses
[1] Shaun Bangay. Modelling Parallel and Distributed Virtual Reality Systems for Performance Analysis and Comparison. PhD thesis, Department of Computer Science, Rhodes University, Grahamstown, South Africa, November 1996. [PDF] [BibTeX]
Images