I work on data corruption issues on Hadoop, using detailed information like Performance Counter and file system operations, on security isolation, on regular expression, and on program diagnosis.

The internship project is to automate cherry-pick discovery and management. The automated process of discovering pairs of commits that introduces break-ages and fix breakages, in other words, a cherry-pick, enabling not only cherry-picks that developers manually putinto the performance regression experiments pipeline but also automatically manage them in databases throughanalyzing the history information of successful and failed experiments in incident trackers. The automation is extended to handle not just one single breakage, but also consider multiple breakage scenarios.

The project measures the importance of mining domain-specific knowledge by identifying string literals in thousands of selectedGitHub projects by leveraging Rosie pattern language, which is an alternative of regular expression language,and Apache Spark for data processing. The analysis of those string literals demonstrated the difficulty of avoidinghard-coding and the security vulnerabilities brought in by hard-coding string literals.

Explored the unsupervised system metrics segmentation and log differentiation, discovered system actions of significant impacts by combining both change point detection for system metrics and trained log templates logs, and comparing the changes of different templates in each segmentation.

I designed an analysis graph for dependency relationships in projects with D3.js, jason and Java Management Tools like maven and ivy