An Efficient Hybrid Approach Based on the ACO and CLA for Static Task-Graph Scheduling in Homogeneous Multiprocessor Environments

Task scheduling has been so far of important challenges in high-performance computers e.g. parallel and distributed systems. Using such architectures during compiling, each application program is divided to some tasks. Because of data-flow among the tasks, they may be dependent to one another; hence, there will be precedence constraints and communication delays among them so that each application with its corresponding tasks can be modeled using a Directed Acyclic Graph (DAG) named task graph. In static task-graph scheduling in homogeneous multiprocessor environments, tasks in the given task graph should be mapped to a predefined number of identical processing elements regarding the precedence constraints and communication delays so that the program’s completion time (finish time) is minimized, and this is an NP-hard problem form the timecomplexity perspective. Actually, the achieved results are dominated by two different-in-nature factors: 1) which topological order of tasks should be considered? (sequencing subproblem), and 2) how should the extracted order be distributed over the processors? (assigning subproblem). In this paper, an efficient hybrid approach is proposed in which the Ant Colony Optimization (ACO) determines the order of tasks, and a Cellular Learning Automata (CLA) machine tackles with the assigning subproblem, and maps the task order derived by ACO to the existing processors. 125 randomly-generated task graphs with different shape parameters such as size, Communication-toComputation Ratio (CCR), and parallelism are used for the comparison study, and the results shows that the proposed approach is more successful than the traditional counterparts from the performance point of view, and eventually outperforms them.