A task mapping simulation framework for comparing the performance of mapping heuristics in various scenarios

Heterogeneous high-computing distributed systems need to process tasks as efficiently as possible by mapping each task to the most suitable machine from the system. Mapping heuristics can be used to solve this problem, but the performance of these heuristics depend on the environment in question. In this paper we propose a highly-customizable Task Mapping Framework for comparing heuristics that can be used in various scenarios based on performance metrics. Our framework was used to test ten mapping heuristics in eight scenarios using four performance metrics: the makespan, the load imbalance, the algorithm´s execution time and the success rate. The tasks used in the simulation had priorities and soft-deadlines, and the scenarios focused on comparing between a low and a high number of tasks, consistent and inconsistent ETC matrices, and a low and a high heterogeneity using a uniform and a gamma random distribution of the tasks´ execution times. This framework proved to be an efficient tool for determining the best mapping heuristics in different scenarios.