I am interested in doing experimental research on large-scale distributed systems, such as clusters, grids, and clouds, with a particular focus on performance and reliability aspects of such systems.

• Contributed to the design of a dynamic resource allocation mechanism that balances the allocations of multiple MapReduce framework instances to provide each instance similar levels of quality of service. (ACM SIGMETRICS'14)
• Designed and analyzed various machine learning-based performance models to automatically tune Hadoop MapReduce applications. (MASCOTS'13, Best Paper Award)
• Contributed to the design of a system to deploy multiple MapReduce framework instances to multi-cluster systems. (MTAGS'12 @ SuperComputing , Best Paper Award)
• Designed and assessed several heuristics for energy-efficient scheduling of Hadoop MapReduce workloads on heterogeneous clusters comprising low-power and high-performance processors. (GCM'11 @ ACM/IFIP/USENIX Middleware )

• Evaluated the variability of the performance delivered by production cloud services using the traces that we collected from Amazon Web Services and Google App Engine. (CCGRID'11 )
• Using scientific computing workloads evaluated the performance of four production clouds: Amazon Elastic Compute Cloud (EC2), Mosso, Elastic Hosts, and GoGrid. (IEEE TPDS , Biggest Impact Award)
• Developed a framework, C-Meter, for running performance tests on IaaS clouds. (Cloud'09 )

• Investigated various overload control techniques to control overload in multi-cluster computer systems with extensive experiments in the DAS-3 multi-cluster grid in the Netherlands. (GRID'11 )
• Investigated the performance of dynamic workflow scheduling policies with both realistic simulations and real system experiments in the DAS-3 multi-cluster grid. (HPDC'10 )
• Investigated the impact of static and dynamic overprovisioning techniques on the performance consistency of Bag-of-Tasks workloads. (Grid'10 )
• Evaluated the performance and benefits of job runtime and queue wait time predictions in grids using traces gathered from various research and production grid environments. (HPDC'09 )

• Proposed a model for groups of failures (space-correlated failures) in large-scale distributed systems using fifteen traces from the Failure Trace Archive. (EuroPar'10 )
• Investigated and proposed a model for time-correlated failures (investigated auto-correlation, periodicity and peaks of failures) in large-scale distributed systems using nineteen traces from the Failure Trace Archive. (Grid'10 )

• This research has been done in the context of the SLIPStream project when I was an Intel/CMU Summer Research Fellow at Intel Research Labs, Pittsburgh.
• Designed and implemented algorithms for dynamic and incremental placement of perception applications to minimize the end to end latency subject to capacity, placement, and migration cost constraints. (HPDC'11 )