SOFTWARE: Building High Performance and Reliable System Infrastructure for Clustered Scientific Data Centers

College Of William And Mary, Williamsburg VA

Investigators

Abstract

Internet Data Centers are built by clustered high performance servers, workstations, and PCs, where the size of the cluster ranges from LAN (local area networks) to WAN (wide area networks for metasystems and for grid computing). The data centers not only provide data searching/storage services, but also provide large scale computing services. Effective resource management presents important and challenging research and development issues to be addressed in the areas of scalability and reliability. Cluster-based data centers provide us with tremendous opportunities for aggregating huge amounts of information and computing resources in a cost-effective way, but it is harder to build distributed data centers that are efficient. This exploratory award will investigate several areas of research in resource management in distributed systems based on the core technology of being able to monitor and predict the usage of resources in a distributed system: (1) development of a unified network and system information extraction tool for performance prediction, and data center resource utilization; (2) modeling of and profiling network behavior and activities; and (3) development of dynamic scheduling schemes and their implementations for optimal utilizations and management of global resources in Internet data centers including networks, memory, and I/O systems. The result will be the identification of the most promising line of research for future investigations.