Poster: Structural Comparison of Parallel Applications
Date: Tuesday, November 19, 2013, 05:15 PM - 07:00 PM
Room: Mile High Pre-Function
Description: With the rising complexity of both architectures and applications, performance analysis and optimization has become essential in the development of scalable applications. Trace-based analysis has proven to be a powerful approach. However, a manual comparison of traces is extremely challenging and time consuming because of the large volume of detailed data and the need to correctly line up trace events. Our solution is a set of techniques that automatically align traces so they can be compared, along with novel metrics that quantify the differences between traces, both in terms of differences in the event stream and timing differences across events. We apply a hierarchical clustering approach that computes a dendrogram containing groups of structurally identical traces by iteratively increasing the number of included call-levels. We demonstrate the effectiveness of our solution by showing automatically detected performance and code differences across different versions of two real-world applications.
Links: Official link from SC13