Runtime Analysis of Distributed Data Processing Programs

Debugging massively parallel data analysis programs is cur-rently a difficult process. Traditional debug cycles involve manual code instrumentations, re-execution and analysis of the resulting data. This is expensive in terms of develop-ment time, execution time, amount of data produced, and cognitive overhead. This work proposes a course of research that is meant to alleviate this situation by automating the code instrumentation and by lowering the re-execution time of instrumented code. By using these techniques, we hope to achieve a higher efficiency compared to manual debugging approaches. 1