Near Optimal Task Graph Scheduling with Priced Timed Automata and Priced Timed Markov Decision Processes

TL;DR

This paper introduces a methodology for near-optimal task graph scheduling using Priced Timed Automata and Priced Timed Markov Decision Processes, reducing complexity and improving schedule quality.

Contribution

It presents a novel reduction of task graph scheduling to location reachability in PTA and PTMDP, with implementation and experimental validation showing improved schedules.

Findings

Many schedules are shorter or equal to the best-known schedules.

Chains reduce computation time effectively.

Method applies to preemptive and non-preemptive scheduling.

Abstract

Task graph scheduling is a relevant problem in computer science with application to diverse real world domains. Task graph scheduling suffers from a combinatorial explosion and thus finding optimal schedulers is a difficult task. In this paper we present a methodology for computing near-optimal preemptive and non-preemptive schedulers for task graphs. The task graph scheduling problem is reduced to location reachability via the fastest path in Priced Timed Automata (PTA) and Priced Timed Markov Decision Processes (PTMDP). Additionally, we explore the effect of using chains to reduce the computation time for finding schedules. We have implemented our models in UPPAAL CORA and UPPAAL STRATEGO. We conduct an exhaustive experimental evaluation where we compare our resulting schedules with the best-known schedules of a state of the art tool. A significant number of our resulting schedules are shown to be shorter than or equal to the best-known schedules.