Scheduling algorithms for distributed/parallel computation on Beowulf clusters

Beowulf clusters are proving to be a cost-effective alternative to costly parallel computers. The performance of parallel and distributed programs running on a Beowulf cluster can be improved if the underlying LINUX operating system provides support for coscheduling mechanisms. The goal of coscheduling is to achieve simultaneous scheduling of communicating parallel processes on their respective nodes thereby reducing the communication lags between them. This report discusses some coscheduling mechanisms in general and describes the implementation and evaluation of the Periodic Boost-Decay algorithm in particular. We find that Periodic Boost-Decay algorithm provides significant improvement in the performance of parallel programs run on a Beowulf cluster.