Energy-efficient optimization strategy for simulation task scheduling based on supercomputing platform

To address the problems of high energy consumption and underutilization of resources in the supercomputing platform of the grid center, a regression analysis is used to deal with the relationship between resource usage data and energy consumption, and an energy consumption evaluation model is created. The proposed energy consumption assessment model is combined with the existing simulation task scheduling strategy to propose an energy consumption-aware load balancing strategy to improve the overall resource utilization rate from the perspective of improving the load balancing of computing nodes on the platform; a node adjustment algorithm based on real-time CPU utilization is proposed to dynamically adjust the number of nodes involved in computing. The task-node mapping relationship is scheduled according to the load of computing nodes to reduce the energy consumption of the platform without significant impact on the execution time of existing power simulation tasks. The energy consumption modeling and performance testing of the scheduling scheme are carried out under Linux system and through simulated simulation environment such as CloudSim. The experimental results show that the energy consumption evaluation model based on the system resource utilization rate can achieve about 95% accuracy; the total energy consumption of the two scheduling strategies is basically the same, and there is about 16% energy saving compared with the basic scheduling strategy and about 11% energy saving compared with the Max-Min scheduling strategy.