Elasticity of Workloads and Periods of Parallel Real-Time Tasks

Abstract

The elastic task model allows sequential periodic real-time tasks, such as those found in multimedia players and adaptive control systems, to adjust their periods dynamically to manage quality of service or to accommodate other tasks. Recent theoretical advances show that parallel real-time tasks can adapt their periods similarly. This paper further extends the concept of elasticity of parallel real-time tasks, to allow them to adapt their computational workloads instead of their periods, such as when a real-time video processing application can improve image quality if it can do more computation within a given period. This paper also presents a new concurrency platform in which each parallel real-time task can adapt either its period or its workload, supporting heterogeneous forms of elasticity for different application needs. Empirical evaluations we have conducted (1) demonstrate the ability of this concurrency platform to enforce theoretical guarantees from both prior work and results developed in this paper, and (2) help to quantify and demonstrate tradeoffs between temporal and computational elasticity.