Inherently stable priority list scheduling in systems with external delays

Abstract

Low-overhead solutions to the problem of scheduling instability in nonpreemptive priority-list task scheduling for real-time systems are presented. Scheduling instability occurs in precedence-constrained task systems when a reduction in the duration of one or more tasks causes the start of a subsequent task to be delayed, so that real-time deadlines could be missed. A recent scheduling model introduced phantom tasks into the precedence graph to model processes and events external to the processors. This model is exploited to derive new, inherently stable run-time task dispatchers of varying complexity. Performance simulations showed that for a wide range of task systems, simple dispatchers performed remarkably well compared to more complex dispatchers. The implications are that nonpreemptive list scheduling can be stable, efficient, and low in terms of run-time overhead. >