Contributed to Spark, an open source cluster computing framework written in Scala. Advised by Matei Zaharia and Scott Shenker.

• Led the design and development of Arthur, a distributed replay debugger for Spark programs; gave a talk at the AMPLab Winter 2012 retreat (poster, slides); and wrote a technical report [3]
• Used Spark to build Bagel, a Pregel-like graph processing framework, and presented a poster at the AMPLab Summer 2011 retreat
• Coauthored an NSDI 2012 paper [1] on Spark that won Best Paper Award, contributing Bagel and an evaluation of Spark's user-controllable partitioning

Built CloudClustering, a scalable clustering algorithm on the Windows Azure cloud, for Microsoft Research's eXtreme Computing Group using C#. Generalized the design into a set of architectural patterns for data processing using cloud services. Published a workshop paper [2] and gave a talk at DataCloud 2011 (slides).

Contributed to Google's workflow execution system, which handles collections of processes with dependencies.

• Designed and implemented an algorithm to improve efficiency at allocating cluster resources for complex workflows
• Productionized a tool for static analysis of user workflows to help design and prioritize new features
• Contributed cache-related bugfixes and features to Google's next-generation data processing substrate

Added network usage tracking and limit enforcement to Facebook's cluster manager using C++ and Linux cgroups.