Multi-Objective Scheduling Algorithm for Heterogeneous GPU Clusters: Optimizing Completion Time and Energy Consumption

This paper presents a multi-objective scheduling algorithm for heterogeneous GPU clusters. The algorithm, based on genetic algorithms and simulated annealing, aims to simultaneously optimize task completion time and energy consumption. Specifically, the algorithm maps tasks to GPU nodes and schedules them based on the tasks' computational and communication requirements and the energy characteristics of the GPU nodes. Subsequently, the algorithm utilizes genetic algorithms and simulated annealing to optimize the total task completion time and total energy consumption. Experimental results indicate that the algorithm achieves superior performance and energy efficiency on heterogeneous GPU clusters.