The design and implementation of a decentralized scheduling facility for a workstation cluster

A decentralized scheduling facility designed for a large, heterogeneous, workstation-based distributed system is described. A system with existing facilities for remote execution and (possibly) task migration is assumed. The solution proposed is not only performance-effective but also robust against processor and communication failures, scalable to large systems with a high frequency of scheduling needs, and stable. It has been implemented on a large distributed system based on the V kernel. Several scheduling requirements of parallel computations, executing as task groups where the individual tasks executed in parallel on different workstations, are exposed. These requirements significantly complicate the scheduling process. A possible solution for effectively scheduling these task groups is presented