A Node Access Frequency based Graph Partitioning Technique for Efficient Dynamic Dependency Analysis

Program execution traces (simply "traces" for short) contain data/control dependency information, and are indispensable to novel kinds of debugging such as back-in-time debugging. However, traces easily become large and complicated. For a practical use, maintainers need to be able to interactively invoke an analysis process when required and obtain rapid feedback. To this end, the authors develop an approach for efficient macroscopic analysis of traces of large sizes with complex data structures. We propose an approach that involves storing graphs in a database that reduces the number of attributes in the main memory during dependency analysis. We also introduce a criterion for the application of this approach that can maximize its effectiveness. Finally, we conduct experiments to assess its effectiveness for efficient dependency analysis.