A power consumption prediction algorithm based on job similarity judgment
Meihong Yang, Yongjie Chen, Jibin Wang, Ying Guo, Shuwei Qin · 2023
Facing the high energy consumption problem of High Performance Computing(HPC) clusters, power consumption prediction of HPC clusters can help data centers improve energy efficiency and formulate energy-saving strategies. However, the existing power prediction methods predict the power consumption of many different jobs in an HPC cluster, and due to the variety of job types in the cluster, the power consumption of different types of jobs varies, and even the same job may change due to different parameters, thus reducing the accuracy of the prediction. To address the current problem of low accuracy of power consumption prediction for multiple jobs in HPC clusters, we propose a power consumption prediction method(PCSSJ). based on the similarity judgment of user jobs The algorithm consists of two modules: similarity judgment and power consumption prediction. The algorithm can select a set of jobs similar to the target job from the historical jobs executed by this user for training and predict the target job. To verify the effectiveness of the PCPSJ algorithm, we train on five different types of datasets and compare them with the baseline model, and the results show the superiority of the PCPSJ algorithm.