Guaranteeing deadlines in MAFT

Abstract

The author discusses scheduling in the multicomputer architecture for fault-tolerance (MAFT), a distributed system designed to provide extremely reliable computation in real-time control systems. Scheduling in MAFT is based on a fault-tolerant variation of a priority-list schedule. The threading delay associated with agreement on task completions can cause schedules to experience Richards' anomalies, in which legitimate run-time behavior can cause tasks to miss deadlines. The author develops theory to stabilize the fault-tolerant priority list. She gives an algorithm that stabilizes a restricted class of task workloads and discusses extensions of her results to other workloads.<>