On the periodic behavior of real-time schedulers on identical multiprocessor platforms

TL;DR

This paper establishes a general periodicity upper bound for any deterministic, memoryless real-time scheduling algorithm on identical multiprocessor platforms, regardless of task constraints or hardware architecture.

Contribution

It introduces a universal periodicity result that applies broadly to various scheduling scenarios, providing an upper bound based solely on task releases and deadlines.

Findings

Provides a general periodicity upper bound for feasible schedules.

Applicable to any deterministic, memoryless scheduler, including non-work-conserving algorithms.

Independent of specific task parameters beyond release times and deadlines.

Abstract

This paper is proposing a general periodicity result concerning any deterministic and memoryless scheduling algorithm (including non-work-conserving algorithms), for any context, on identical multiprocessor platforms. By context we mean the hardware architecture (uniprocessor, multicore), as well as task constraints like critical sections, precedence constraints, self-suspension, etc. Since the result is based only on the releases and deadlines, it is independent from any other parameter. Note that we do not claim that the given interval is minimal, but it is an upper bound for any cycle of any feasible schedule provided by any deterministic and memoryless scheduler.