I am a Ph.D. student at Cornell University being advised by Professor Rachit Agarwal.
My research interests include distributed systems, data-center networking and peer-to-peer systems.
My research focuses on two areas:
Resource disaggregation in datacenters. This involves augmenting the research communities understanding of how systems change when moving to disaggregated architectures and building systems that can take advantage of the elasticity and fine-grained resource multiplexing that disaggregation enables without degraded performance.
Privacy-aware distributed systems for the cloud. Systems run on public clouds leak information (network, storage, memory) accesses that can be used to learn abusable information about the data being accessed. Initial work in this area characterizes the information leakage of existing encrypted key-value stores and involves building a high-performance key-value store whose leakage cannot be abused.
Worked with the Network Infrastructure team in Sunnyvale, CA to to improve network bandwidth allocation on Google's inter-datacenter WAN. This work builds upon Google's hierarchal bandwidth allocator, BwE. It involved developing new bandwidth allocation techniques that leveraged machine learning and real-time control literature; and algorithmic analysis of existing bandwidth allocation techniques.
Worked with the Azure Data Warehouse team in Aliso Viejo, CA to build a prototype for a distributed query store for Azure SQL Data Warehouse. This enabled more extensive query monitoring and debugging for distributed queries, especially in the case of complex query execution plans spanning many servers.
Researched techniques for realizing latency-sensitive high performance computing applications in less predictable and higher latency environments, such as commercial clouds. This involved optimizing application communication and minimizing latency introduced by the runtime system through lock-free algorithms. Other work included designing a RDMA-backed message passing interface for CRAY machines, and building parallel object replication techniques for load balancing.
Worked to optimize a parallel program runtime system Charm++, a parallel object-oriented programming language, for performance. Additionally, modified parallel algorithms for shared memory parallel applications to operate scalably in a distributed memory context.
Worked with the Google Compute Engine team to improve cloud network reliability for development clusters through developing network failure detection tools.
University of Illinois at Urbana-Champaign
Documents and Contact