We live in an exciting time for systems research given the explosion of digital data mined from our world and social relationships, new experiences such as cloud computing and the mobile Internet, and incredible advances in computing devices, storage media, and connectivity. We engage in fundamental systems research with expertise that spans theory and practice in distributed systems, storage systems, cloud computing, networking, and programming languages. Over the years, we have designed, built, analyzed, and optimized various large-scale production distributed systems that power on-line services. These experiences have inspired a series of research projects, producing significant results in top systems conferences such as SOSP, OSDI, NSDI, Eurosys, and VLDB.
- PASS: Program Analysis for SCOPE ScriptsPASS project is a continuing collaboration with the Cosmos team that aims to improve SCOPE script correctness and performance using program analysis techniques, following the inter-disciplinary research direction, among program language, system and database research.
- Temporal Graph Storage and Analysis of Social DataAn explosion of user-generated data from online social networks motivates analysis to extract deep insights from this data's graph of social, temporal, spatial, and topical connections. We are building a system to enable storage and analysis of such graphs that considers their evolution over time as trending topics and social activities change.
- TimeStream: Large-Scale Real-Time Stream Processing in the CloudTimeStream is a distributed system designed specifically for low-latency continuous processing of big streaming data on a large cluster of commodity machines. The unique characteristics of this emerging application domain have led to a significantly different design from the popular MapReduce-style batch data processing. In particular, we advocate a powerful new abstraction called resilient substitution that caters to the specific needs in this new computation model.
- MadLINQ: Large-Scale Distributed Matrix Computation for the CloudThe computation core of many data-intensive applications can be best expressed as matrix computations. The MadLINQ project addresses the following two important research problems: the need for a highly scalable, efficient and fault-tolerant matrix computation system that is also easy to program, and the seamless integration of such specialized execution engines in a general purpose data-parallel computing system.
Tango: Distributed Data Structures over a Shared Log, in SOSP 2013 [pdf]
A Characteristic Study on Failures of Production Distributed Data-Parallel Programs, in ICSE (SEIP track) 2013 (best paper!) [pdf]
Failure Recovery: When the Cure Is Worse Than the Disease, in HotOS 2013 [pdf]
TimeStream: Reliable Stream Computation in the Cloud, in EuroSys 2013 [pdf]
Want to drop by?
12F, Microsoft Building 2, Danling Street, Zhongguancun, Haidian District, Beijing, China
Click here for Map Location