MILP-based Deadline Assignment for End-to-End Flows in Distributed Real-Time Systems

Abstract

End-to-end flows, which have a set of chainlike subtasks, are widely used in distributed real-time systems. For instance, multimedia and automative applications require that subtasks finish executing on a chain of processors before their end-to-end deadlines. The scheduling of such chained subtasks decides the schedulability of a distributed realtime system. Since the subtask priority assignment problem is NP-hard in general, most heuristics are presented to schedule end-to-end flows in two separate steps. The first step calculates intermediate relative deadlines for frames, and the second step makes scheduling decisions under EDF scheduling. Because the quality of the priority assignment of subtasks will directly affect the schedulability of the distributed systems, the two separate steps may cause pessimism in schedulability analysis. To reduce potential pessimism, we combine the two steps in our novel dGMF-PA (distributed generalized multiframe tasks with parameter adaption) model. We present an algorithm based on mixedinteger linear programming for optimally selecting frame relative deadlines in the dGMF-PA model. An approximation algorithm is also proposed to reduce computational running time. Our approximation algorithm has a tunable speed-up factor of 1 + where can be arbitrarily small, with respect to the exact schedulability test of dGMF-PA tasks under EDF scheduling. Extensive experiments have shown that our approximation algorithm (which is a sufficient schedulability test) can schedule at most 44 % more than HOSPA, an existing state-of-the-art algorithm.

Multi-criteria Optimization of Real-time DAGs on Heterogeneous Platforms under P-EDF

“…In DAG-alike models of computation, a problem that received a fair amount of attention is the one of designing deadline splitting strategies, so to be able to divide an end-to-end application deadline into a set of local deadlines to be used for individual real-time tasks [22,59]. This problem was tackled also using MILP-based approaches [53].…”

End-to-End Analysis and Design of a Drone Flight Controller

“…While the real-time community has developed valuable approaches to scheduling and response time analysis of tasks [10], end-to-end timing analysis has received only limited attention. Most of the prior work has originated from real-time network communication research [34,30,28,40,33,42], and is based on event-triggered communication with FIFObased buffers. In a drone flight control program, however, single register-based buffers and periodic sampling are more common.…”