Abstract:
Today's large distributed computing systems have unprecedented scale of size and distribution, which pose great challenges to data and resource management. Virtualization offers unique opportunities to address these challenges by enabling abstractions of resources and services that provide a basis for flexible, dynamic system management and optimization.

This talk will focus on the virtualization of distributed file systems for scalable and flexible grid data management. This approach is novel in that it provides on-demand cross-domain data access without modifying applications and operating systems, and supports application-tailored enhancements on performance, consistency, security, and fault-tolerance. Furthermore, it leverages autonomic services to dynamically manage data provisioning according to the application requirements and changing resource availability. Experiments with typical file system benchmarks and scientific applications show that our approach can substantially outperform traditional distributed file systems in wide-area networks, and automatically optimize data access and recover from failures in a dynamic environment. The talk will be concluded with a vision of future large-scale virtualized systems for high-performance and high-throughput computing.

Biography:
Ming Zhao is a PhD candidate in the Department of Electrical and Computer Engineering at the University of Florida. He received his B.S. and M.S. from Tsinghua University. His research interests include distributed systems, virtualization, grid computing, and autonomic computing. He has actively published in these areas and received the best student paper award at the International Conference on Autonomic Computing 2007.