Research Areas:


Most of my current research is focused on the areas of unified power and cooling management for data centers and data-intensive computing. You can find a brief description of this work below and a more in-depth discussion of it in my publications.

Unified Power and Cooling Management:

My work on energy management looks primarily at the energy-efficiency of data centers. Previous research has optimized either server power or the power used by cooling equipment independently. In contrast, this research looks at a unified approach for workload, power, and cooling management. The underlying assumption in our unified power and cooling management systems is that both the compute and facility systems are aware of the other and can therefore make better-informed decisions. Our systems generally use a model-based approach to predict the impact of their optimizations on both energy costs and performance of hosted workloads. This is an interdisciplinary project and involves researchers with expertise in Mechanical Engineering, Control Theory, Distributed Systems, and Virtualization.

Data-Intensive Computing:

My work within data-intensive computation focuses on the implication of new technologies (such as photonics and non-volatile storage) on "big data" applications that use MapReduce frameworks such as Hadoop. In addition, we are also looking at intelligent resource scheduling of these applications on Infrastructure-as-a-Service (IaaS) systems such as Amazon.com's EC2. We are also working on benchmark creation as a part of this project and we plan on releasing them, along with their corresponding data sets, to the community.

Content Addressable Storage:

As a graduate student, a large fraction of my research focused on the use of Content Addressable Storage to improve the performance of applications over Wide-Area Networks. In particular, I showed how external, and potentially untrusted, resources can be used in an opportunistic manner to optimize bulk data transfer in WAN-based client-server systems. By using content-addressability to detect similarity between different data sources, these techniques allowed external sources to be used without weakening any attributes, including consistency, of legacy systems and with no or minor changes to the original system. My dissertation validated this claim empirically through the use of five case studies that encompassed the two traditional forms of data storage, file systems and database systems, and then generalized the claim in the form of a generic transfer service (DOT) that can be shared by different applications.