A Framework and Methods for Dynamic Scheduling of a Directed Acyclic Graph on Multi-core

Abstract

Abstract—The data flow model is gaining popularity as a programming paradigm for multi-core processors. Efficient scheduling of an application modeled by Directed Acyclic Graph (DAG) is a key issue when performance is very important. DAG represents computational solutions, in which the nodes represent tasks to be executed and edges represent precedence constraints among the tasks. The task scheduling problem in general is a NP-complete problem[2]. Several static scheduling heuristics have been proposed. But the major problem in static list scheduling is the inherent difficulty in exact estimation of task cost and edge cost in a DAG and also its inability to consider and manage with runtime behavior of tasks. This underlines the need for dynamic scheduling of a DAG. This paper presents how in general, dynamic scheduling of a DAG can be done. Also proposes 4 simple methods to perform dynamic scheduling of a DAG. These methods have been simulated and experimented using a representative set of DAG structured computations from both synthetic and real problems. The proposed dynamic scheduler performance is found to be in comparable with that of static scheduling methods. The performance comparison of the proposed dynamic scheduling methods is also carried out.