06/01/1992 09:00 AM
Computer Engineering
Distributed operating systems give users access to multiple computational engines throughout the system network. Users of one workstation are not hindered by the CPU intensive applications run on a different workstation. However, when a large number of machines in the network are idle, the efficiency of computation is decreased. Load balancing promises to alleviate this problem by sharing the workload on heavily loaded workstations with lightly loaded workstations. A threshold can be established so that only a limited number of processes are transferred to the lightly loaded sites, so that the load at those sites does not inhibit local execution of processes. While load balancing has been studied extensively, there have been few studies on determining which processes should be relocated to increase the throughput of the system. We present an automatic process selector for load balancing that chooses processes to be relocated based on the history of its performance. We have implemented this system on the Sprite distributed operating system. Notes: M.S. Thesis