Abstract

It is important to understand performance aspects of a computer system from the software architecture and its configurations. For a system that has various possible configurations, the performance analysis will become difficult. This thesis describes a compositional model-building approach which is useful when there are many possible configurations. This approach is based on assembling the sub models and is demonstrated in a network software router, CGNet.

In addition, the derivation of parameters must be well addressed for performance modeling. The measurements and parameter estimation techniques can be deployed in completing the performance model. We apply these approaches and techniques to discussing the real problems we encountered in parameterizing the performance model.

The converter tool, which automates the compositional model-building approach, has been developed for CGNet. This tool is a solution to bridging the gap between the network configurations and the performance analysis. It is a configuration-based model-building tool for CGNet. With the tool, we can generate the performance model from configurations.

-- Daniel Amyot - 11 Aug 2006

Discussion

• Please feel free to discuss this article directly on this page. Constructive comments are welcomed! Please sign your TWiki name.