Distributed Resource Management in Workstation and PC Clusters

点击翻译按钮获取中文摘要

Advances in network and processor technology are greatly changing the communication and computational power of local-area clusters of workstations and personal computers. In particular, the replacement of Ethernets by newer switched network technologies, which provide both higher throughput and lower latency, presents an opportunity for operating systems to better use network resources. This research seeks to improve the performance of applications through distributed (cluster-wide) resource management. The project will develop algorithms that permit the effective sharing of all network resources: processors, network, memory, and disk storage. For example, programs on heavily- loaded nodes can use available memory on idle nodes for both file system and virtual memory backing store, rather than disk, thereby reducing the data latency on access to that data. As another example, applications may migrate to nodes that provide either a better processing environment or a better memory environment for their execution. A key issue in developing algorithms for such closely-coupled system is finding the right balance in order to optimize system-wide performance while minimizing any degradation of performance on individual systems as a result of network-wide resource sharing. ***