PARAdeg-processor scheduling for acyclic SWITCH-less program nets

• Published in: Journal of the Franklin Institute Vol. 336; no. 7; pp. 1135 - 1153
• Main Author: Ge, Qi-Wei
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.09.1999
• Subjects: Data-flow program; Evaluation; GA scheduling; List scheduling; Multiprocessor scheduling; Program net; Multiprocessor scheduling; Evaluation; Data-flow program; List scheduling; GA scheduling; Program net
• ISSN: 0016-0032; 1879-2693
• DOI: 10.1016/S0016-0032(99)00026-5

This paper deals with multiprocessor scheduling for acyclic SWITCH-less program nets that is a graph representation of a data-flow program. A task graph is a special case of acyclic SWITCH-less net PN and the important difference is that a program net allows nodes to fire more than once while a task graph asks its each node to fire exactly once. Hence scheduling problem for program nets is NP-hard in a stronger sense. In this paper, we first extend CP (critical path) method by taking firing numbers of nodes into account to propose extended critical path (ECP) method. Based on this we further propose ORF/ECP (OR-node First/ECP), and LTF/ECP (Larger node firing Time First/ECP) methods, by taking notice of firings of OR-nodes and node firing times, respectively. Then using priority lists as populations, we propose two genetic algorithms n-GA and 1-GA, by, respectively, requiring all and only initial populations satisfying precedence relation of nodes. Our experiments show that: (i) precedence relation of nodes is no more important for scheduling of program nets; (ii) 1-GA gives the best results, and LTF/ECP is next to 1-GA and is effective when node firing numbers are not large.