Compiler and Runtime Support for X10 Work-Stealing (Jan. 2010 - Present)
X10 (http://x10-lang.org/) is a new programming language being developed at IBM Research in collaboration with academic partners. X10 is a type-safe, parallel object-oriented language, and it's also a member of the Partitioned Global Address Space (PGAS) family of languages. We are developing the Work-Stealing scheduling support in X10 compiler and runtime. The goal is to improve X10's performance and scalability on fine granularity parallel tasks.
Java Performance Tool on Multi-core Platform (Dec.2006 - Dec.2009)
In the multi-core/massive parallel computing era, Java based middleware and applications should be optimized to utilize the underlying hardware computation power. This type of performance analysis needs collecting data from different layers, including the application, middleware, OS and hardware platform. However, most state-of-art Java performance tools only focus upper layers. We developed a new Java performance tool, which bases on vertical profiling, and makes use of advanced data processing and visualization techniques to help users identify possible performance bottlenecks of Java applications on multi-core platform.
Performance Analysis and Tuning of Websphere SIP Server (Feb. 2007 - Jan. 2008)
SIP (Session Initiation Protocol) provides an open, standard, flexible framework for building a variety of communication services, and it has been widely adopted by the telecommunications industry. IBM Websphere Application Server has the built-in support for SIP. However, the previous SIP component cannot scale up on servers with massive multi-threads. We studied the SIP component in Websphere, and identified the hot lock and Java memory bloat problems in Websphere SIP. We re-implemented some key components in the SIP module, and achieved around 7x performance gain on typical multi-core platforms.
Performance Analysis and Visualization Tool for IBM HPC (Aug. 2006 - Dec. 2006)
Performance Explorer is a performance data analysis and visualization tool for IBM HPC. In this project, we improved its capability and added some features for analyzing MPI trace data better. We added a MPI trace preprocessing component and several new visualization views to calculate and illustrate the communication time and computation time across different computation nodes.
Asset Based Business Transformation Methodologies and Tools (Jan. 2005 - Dec. 2005)
We analyzed reusable assets and reuse patterns in enterprises, and defined an enterprise asset model according to OMG RAS (Reusable Asset Specification). Then we developed an asset repository POC to harvest, manage and reuse assets in the lifecycle of business transformation. The POC was a typical three-tier J2EE application on IBM Websphere Application Server. Two types of clients were also implemented, including a portal application client on Websphere Portal Server and a rich client application based on Eclipse.
Model Driven Business Transformation (Jul. 2004 - Apr. 2005)
The project applied MDA (Model Driven Development) to enterprise IT transformation, which could fasten Business to IT decision and facilitate business process monitoring. We designed the model of MDBT (Model Driven Business Transformation), and developed the MDBT modeling tool based on Eclipse/EMF/GEF. We applied our methodology and tool to a core banking system transformation project of Bank XXX. We also studied the COBOL based legacy system, and analyzed its data flow and control flow.
Projects in Graduate School Period
Biometrics Based Identity Recognition (Jan. 2002 - July. 2004)
It was a key project of the Science and Technology Commission of Shanghai Municipality, and its goal was to create the intellectual properties and a real system to recognize human identity based different biometrics. My research focused on the face recognition channel, including the algorithm and system for face detection and recognition in real-time video stream.
Coal Component Analysis System of Bao Steel (Jul. 2001 - Jan. 2002)
We developed the algorithm and system to detect the ingredients of coal based the digital images captured in the coal transportation system. We applied digital image processing and pattern recognition techniques to analyze the coal image, and received over 90% precision rate.