A multiobjective evolutionary algorithm for QoS-aware planning in heterogeneous computing systems

This article presents the application of a parallel evolutionary algorithm for solving a multiobjective version of the task scheduling problem in heterogeneous computing infrastructures (cluster and grid systems). In real-life scenarios, the scheduling problem must take into account the needs of both service providers and users. Thus, the multiobjective version of the problem solved in this article is relevant to find schedules with accurate trade-off values between the quality-of-service levels (given by deadlines for the tasks) and minimizing the execution time required for a set of tasks submitted to the system. The problem is studied over scenarios with dimensions that represent realistic nowadaus computing infrastructures, and a parallel evolutionary algorithm is introduced to efficiently solve the problem. The experimental analysis considering both problem objectives demonstrate that the proposed algorithm is able to compute high-quality solutions for the problem, with accurate trade-off values between system utilization and quality of service, outperforming a set of well-known deterministic heuristics for hterogeneous computing scheduling.