Developing an Open Architecture for Performance Data Mining

Performance analysis of high performance systems is a difficult task. Current tools have proven successful in analysis tasks but their implementation is limited in several respects. Closed architectures, predefined analysis and views, and specific platforms account for these limitations. Embedded systems are particularly affected by these concerns. This paper presents an open architecture for performance data mining that addresses these limitations. Comparisons of the architecture with current tools show its capabilities address a wider range of system phases and environments.