I am an Assistant Professor in Department of Computer Science at Cornell University.
My research interests are in systems and networking. I am also interested in theoretical problems arising out of building practical systems. My research thus spans (and integrates) systems, networks, and theory.
- Networks that never drop packets: One of the core problems of current network fabrics is packet drops, that directly or indirectly lead to inefficiency and complexity (e.g., high CPU utilization, high latency, complex network stacks). We are designing and building network fabrics and stacks that guarantee that packets will never be dropped.
- Resource Disaggregation: Traditionally, servers have been designed with certain amount of CPU, memory and storage, all on the same motherboard. However, the end of Dennard's scaling and slowdown of Moore's law has led to surfacing of several fundamental limitations of such server-centric architectures. The solution is to decouple compute from storage! To achieve this goal, we are working along several directions:
$3M award from NSF, Google faculty research award.
- Operating System for Disaggregated Architectures
- Network Fabric and Stacks [OSDI'16] [NSDI'19]
- Secure Disaggregated Storage [OSDI'18]
- Storage Stack for Disaggregated Architectures
- Extending Serverless Architectures for Stateful Functions
- Near-Perfect Datacenter Transport Design: Current network datacenter transport designs are extremely heuristic-y, which leads to extremely bad worst-case scenarios. We are designing and building datacenter transport designs that provide provable worst-case guarantees, including:
SIGCOMM'18 Best Student Paper Award.
- Sincronia [SIGCOMM'18] for Coflows;
- Universal Packet Scheduling [NSDI'16] for flexible packet scheduling;
- And a few upcoming surprises ....
- PathDump, an end-to-end system for datacenter network monitoring and debugging. PathDump pushes as much functionality to the end-hosts as possible, while keeping the network fabric as simple as possible [OSDI'16]. The follow-up work enables fine-grained visibility into the network [NSDI'18] and extremely efficient end-host stacks for low-overhead monitoring and debugging [NSDI'19].
Few Past Projects
Here are some of the projects that I have worked on in the past:
- Succinct, a distributed storage system that enables random access, substring search, and even regular expression matches directly on a compressed representation of the input data [NSDI'15]. Follow-up work on BlowFish [NSDI'16] that enables a smooth performance-storage tradeoff, and ZipG [SIGMOD'17] that extends Succinct for very expressive graph queries directly on compressed graphs. Open-sourced and deployed in the real-world.
- Anteater, one of the first systems that proposed the idea of network data plane debugging [SIGCOMM'11]. Laid the foundation for research on network debugging at the data plane.
- Approximate Distance Oracles and Compact Routing, that introduced new data structures, algorithms and techniques for approximate queries on graphs [INFOCOM'11, PODC'13, SODA'13]. Follow-up work led to scalability limits of BGP [ToN'14] and new connections to combinatorial boolean matrix multiplication [ESA'14]. This project led to the first improvement over several classical decade-old theory results.
|SIGMETRICS Tutorials Chair
|NSDI Poster Chair
|HotOS General Chair
- Program Committee