Exact Response-Time Bounds of Periodic DAG Tasks under Server-Based Global Scheduling

Abstract

Artificial-intelligence (AI) techniques are revolutionizing modern safety-critical real-time systems by enabling autonomous features never seen before. However, AI-based workloads are typically expressed as processing graphs that are subject to complex tradeoffs involving parallelism and dataflow dependencies. Due to such complexities, exact analysis of graph-based tasks is challenging under most (if not all) schedulers. This paper presents a periodic server-based scheduling policy for periodic graph-based task systems and provides an exact response-time analysis under this policy. This analysis entails pseudo-polynomial time complexity for pseudo-harmonic periodic graph-based tasks, which are commonly used in practice.