Scheduling jobs that are subject to failure propagation

A single machine is available to process a collection of stochastic tasks. Processing is interrupted when the machine breaks down. The authors introduce a new model of breakdowns that more realistically incorporates the effects that job processing may have on the machine. This failure propagation model is equivalent to a Bayesian formulation in which learning about breakdown rates occurs as jobs evolve. Optimal scheduling policies are described in terms of Gittins indices and these indices are characterised in two special cases. For example, the authors obtain conditions which ensure that an optimal policy will only preempt a job’s processing either at its completion or at a machine breakdown. They also bound the value lost by simplistic modelling which ignores the learning phenomenon.