TMSO: Task Merging Strategies Based Optimized Scheduling for Energy-Constrained Parallel Applications in Heterogeneous Distributed Systems

Currently, distributed computing has become an effective approach to meet the demand for computing power while optimizing the scheduling length under priority constraints. However, the scheduling of parallel application tasks in distributed systems remains inadequately addressed. To solve this problem, this paper proposes a novel scheduling algorithm TMSO for energy-constrained parallel applications in heterogeneous distributed systems. The algorithm integrates HEFT, ESECC and TSSA priority assignment schemes to assign task priorities, merges tasks that meet the preset merging conditions, and optimizes the scheduling process based on the average energy consumption of tasks on different types of processors. For scenarios where no tasks satisfy the basic merging conditions, a critical-path-first lightweight node merging strategy is further proposed, which effectively makes up for the scenario coverage limitation of traditional merging strategies and further reduces the scheduling length. Experimental results show that the proposed TMSO algorithm can effectively shorten the scheduling length of parallel applications under energy constraints; in particular, for applications based on Gaussian elimination transformation, the algorithm reduces the scheduling length by 41.3% compared with the MSLECC algorithm.