Professor; Department of Computer Science and Technology, Tsinghua University
Since as early as 2001, Zheng has been the torchbearer for HPC development in the PRC. His pioneering work in applying new technologies has placed him soundly in the unique position of being both pathfinder and trailblazer, and he’s been very generous in sharing his experiences. A leading advocate and enabler for HPC, he is also a champion of the democratization of HPC, continually working to assure maximum value from the technology as it’s deployed. Now that his work developing the framework for real world applications-based system evaluation and valuation comes to fruition in 2013, it will be interesting to see what he has in store for us in the year ahead.
Zheng is also credited with leading the development of Tsinghua Discovery Series Computer Clusters, which have been applied to many fields such as weather forecasting and network information security, where he gained practical experience in cluster design and assessment. He worked to design a user-level file system to support consistent file states for checkpoint and restart, and the Open MP Compiler, selected by HP and AMD as the compiler of choice for their product lines, and preferred by multiple foreign universities as the platform for conducting their research. Zheng also masterminded the development of a suite of cluster assessment techniques that cover the full system, and are able to identify various system faults caused by hardware, software installation, system configuration and driver mismatch.
Zheng led the development of THUMP107, an embedded CPU running at the highest frequency in China at that time. His work was the foundation for a new Network on Chip (NoC) architecture in which NoC, instead of CPU, maintains the consistency of cache. With computing and communication separated, this architecture eases the design of multicore CPUs and enhances the scalability of CPU.
His method for deploying grid services dynamically and remotely without restarting the service container was adopted by Globus, and he led the development of Network Computing and Application System for Bioinformatics. This work led to the development of a community-based data sharing and storage technology, automatic on-demand software deployment and quick virtual cluster construction. The result was Tsinghua Cloud, the first cloud computing platform in China with more than 15,000 registered users and 400 communities.
He helped to develop the Tsinghua Mass Storage Network System (TH-MSNS), which has neared 100 successful deployments to date in the areas of audit, public security, telecommunication, and education. The university also introduced a coding scheme named GRID, regarded as one of the most classic erasure codes by IEEE Fellow Darrell D.E. Long. This research led to structure-independent techniques for the fast disaster recovery of data centers, which greatly reduces the investment for disaster recovery, improving recovery speed at the same time.
Zheng is the recipient of multiple awards, including the National Award for Science and Technology Progress, First Class-Information Security Management System for Ministry of Information Industry (2002); National Award for Science and Technology Progress, Second Class-High-Performance Cluster and Mass Storage System (2007); National Award for Science and Technology Progress, Second Class-China Education and Research Grid (2008); Best Academic Course Award by Ministry of Education-Computer Architecture (2008); Science and Technology Progress Award by Ministry of Education, First Class-High-Performance Computer Assessment Techniques (2009).
Weimin’s Top 5 HPC initiatives or technologies to watch in 2013:
- The war of many-core accelerators – GPUs have enjoyed great success in the past years as many-core accelerators of HPC systems. With Intel’s release of its Phi processor it will be really interesting to watch to see who will be the winner of this accelerators war.
- Low latency 10GbE vs. Infiniband – InfiniBand has dominated interconnect technology in HPC systems. However, the advent of low latency 10GbE interconnect is changing the landscape of interconnect. Will the low latency 10GbE win a significant portion of small-to-medium scale HPC systems in 2013?
- HPC Cloud – It is not a new idea to execute HPC applications in cloud which has obvious cost benefits. Combined with the low latency 10GbE technologies, we expect that HPC cloud will provide better communication performance to support more HPC applications in the coming year.
- HPC systems built with processors designed in China – China has an impressive record in recent years in Top 500 list. The Tianhe-1A was ranked No. 1 in Nov. 2010, which uses Intel Xeon processors and NVIDIA GPUs and interconnects designed in China. The Sunway Blue Light was ranked No. 14th in Nov. 2011, with processors designed in China. The question is, when will China will have a No. 1 computer in the Top 500 list with processors designed in China? Will it be 2013?
- Next steps to achieving Exascale – The Linpack battle is continuing. We have Titan whose Linpack performance is 17.59 Petaflops. Will we have a machine whose Linpack performance can exceed 50 Petaflops in 2013?