Distributed task force scheduling in multi-microcomputer networks

Abstract

Efficient task scheduling techniques are needed for microcomputer networks to be used as general purpose computers. The Wave Scheduling technique, developed for the MICRO-NET network computer, co-schedules groups of related tasks onto available network nodes. Scheduling managers are distributed over a logical control hierarchy. They subdivide requests for groups of free worker nodes and send waves of requests towards the leaves of the control hierarchy, where all workers are located. Because requests from different managers compete for workers, a manager may have to try a few times to schedule a task force. Each task force manager actually requests slightly more workers than it really needs. It computes a request size which minimizes expected scheduling overhead, as measured by total idle time in worker nodes. Using a Markov queueing model, it is shown that Wave Scheduling in a network of microcomputers is almost as efficient as centralized scheduling.