I am a researcher at the Systems Research Group in Microsoft Research Asia. Previously, I was affiliated with the Wireless and Networking Group in MSRA from 2004 to 2010. I received my Ph.D. in Computer Science from Nanjing University.
My research interests include distributed systems, mobile systems, Internet architecture and protocols, and wireless media communications. I built a few distributed graph engines, among other things.
- Open platform for AI (OpenPAI) (2017 -)
- High performance graph analytics engine (2014 – 2016)
We build GraM, a RDMA-based distributed graph engine that often outperforms exiting solutions by one or two order(s). GraM has helped Microsoft to gain insights in various commercial graph datasets.
- Temporal Graph Storage and Analysis of Social Data (2011 – 2015)
An 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 build systems like Chronos and Kineograph to enable storage and analysis of such graphs that considers their evolution over time as trending topics and social activities change.
- Arming Cloud Services with Task Aspects (2014 – 2016)
Tools help cloud services: model checkers to verify, fault injection to find bugs, replay to debug, and many more. Unfortunately, currently tools are either tediously tangled into service implementations or integrated transparently in ways that fail to effectively capture the service’s problematic non-deterministic behavior. This project makes tooling a first-class concern by having services encoded with tasks whose interactions reliably capture all non-deterministic behavior needed by tools. We show how task aspects can be used to ease the development of an online production data service that runs on hundreds of machines
- Cosmos Custom Job Scheduler Framework (2013 – 2014)
In 2013 I had the privilege to work with the Cosmos team to design, implement, and deploy a custom job scheduler framework that enables the Cosmos runtime to seamlessly integrate third-party job schedulers
- Profiling Workloads in the Bing Data Mining Team (2012 – 2013)
From late 2012 to early 2013 I collaborated with the Bing Data Mining team to profile their queries to understand the major reasons that slow down the query executions and propose improvement methods. Through the profiling, we further established a process to identify “hot” and “cold” data so the team can decide whether to build some index to speedup the queries to “hot” data and retire indices for data that becomes “cold.”
- You and your research. A talk by Dr. Richard Hamming
- “It is slightly better to be simple than correct”. My favorite quote from Worse is Better by Richard P. Gabriel
- An evaluation of the ninth SOSP submissions or how (and how not) to write a good systems paper. Roy Levin and David D. Redell