Scheduling Architecturally-Supported Regions with Precedence-Based Priorities

Abstract

Multicore processor technology provides new possibilities in the domain of cluster-based systems. One of such viable architectural issues are globally interconnected systems of multicore executive modules. The multicore modules can be general purpose or can be architecturally supported to enhance performance for special kinds of functions such as for example included in parallel numerical libraries. The paper is concerned with scheduling methods for modular multicore systems based on global interconnections. A new scheduling algorithm is reported which enables scheduling programs represented by macro data flow graphs to a system of globally interconnected general purpose and architecturally supported modules. The algorithm applies heuristics based on weighted activation graphs of architecturally supported program regions. Experimental results show the advantages of the proposed approach comparing simpler scheduling algorithms based only on topological graph properties.