Author: Andy Bond
Source: GZipped PostScript (0kb); Adobe PDF (114kb)
A network of UNIX [Ritchie, 1974] workstations is becoming a common sight in modern computing environments. Each workstation provides powerful computing resources which are periodically in strong demand by the local user. However even in busy environments, a significant proportion of these machines will be idle or underutilized at any one time. By supplementing the local computing resources through offloading tasks to idle workstations, better utilization can be made of the distributed computing environment. Such a strategy is commonly known as load sharing or load balancing.
This paper discusses an experimental task allocation system called STARS. Using a distributed database of resource usage measurements, STARS allocates tasks to systems based on the availability of resources within the network. We will look at how such a system can be integrated into a user's computing environment and some results based on our experience with its use.