I graduated with a PhD from the Computer Science department at the University of Illinois at Urbana-Champaign. My adviser is Prof. William H. Sanders, and I am a member of the PERFORM group.

Resume (May 2009 version): pdf txt

Research The main focus of my dissertation is to build a framework and develop techniques that would enable an automated approach to evaluate system models and perform optimization on a specific class of models: models whose set of input decision parameters can both be qualitative (non-parametric) and quantitative (parametric) decision variables. To perform a thorough analysis of a large number of configurations with varying system design parameter values, I developed an efficient and algorithmically optimized stochastic simulation technique called simultaneous simulation of alternate system configuration (SSASC), that can evaluate system configurations quickly and accurately. Using this technique, one can evaluate system models 10–100 times efficiently compared to standard simulation algorithms. To evaluate a real case study, We developed and analyzed a stochastic system model of Lustre, a cluster file system (CFS). The models use failure rates computed from real log data, to predict the reliability and availability of the storage architecture of the ABE supercomputer cluster at the National Center for Supercomputing Applications (NCSA).

In addition, I have collaborated with folks from HP Labs to evaluate and design dependable storage systems for enterprize applications in an attempt to solve a real industry problem. I developed a principled, automated approach to designing dependable data storage systems for multi-application environments, which minimized the overall cost of the system while meeting business requirements. My contributions include search heuristics for intelligently exploring the very large design space, as well as modeling techniques for capturing interactions between applications during recovery.

Contact Information