Overview on Job Running Times Prediction Algorithms for HPC Platform

The rapid development of high-performance computers has brought about a tremendous increase in computing power, but the data that needs to be processed is also multiplying. Improving the utilization of computing resources is a fundamental goal of high-performance computing systems. In high-performance computing systems, job scheduling systems often use backfill scheduling strategies to schedule jobs. This strategy is sensitive to the job running time. In the past, users’ job running time was provided by users but was wildly inaccurate, which seriously affected computing resources. Therefore, it is necessary to improve the accuracy of job running time prediction. This article will introduce several types of existing job times prediction algorithms, but they all have some shortcomings. To improve the utilization of computing resources, the prediction algorithm should have high prediction accuracy, low underestimation, and more extensive scope of application.