Use of an Observationally-based Performance Model for Informing Scheduling DecisionsNeil Davies, Use of an Observationally-based Performance Model for Informing Scheduling Decisions. CSTR-94-11, Department of Computer Science, University of Bristol. September 1994. PDF, 134 Kbytes.
Scheduling within a distributed memory computer system requires knowledge beyond the confines of the local processor. Usually this knowledge is exchanged between management entities on the individual processors. This paper describes how a performance model, combined with local observations, can be used to infer the loading of other processors in the systems. The model does not require any additional messages to be exchanged over and above those already being used to support the distributed computation. The performance model is rich enough to capture the effects of multi-threading as well as communications delay. In addition to being used to infer current performance characteristics of the system the model could also be used to predict the likely effects of a scheduling change on the overall performance of the system.