Long Short-Term Memory Networks for Server Workload Prediction

Abstract

The increasing utilization of data centers across generations, fueled by the automation and digitization of numerous services, emphasizes the crucial importance of workload estimation. As these centers play an increasingly central role, optimizing resource usage becomes paramount. This paper explores the critical aspect of workload prediction and its impact on efficiency and costs within data centers. The study examines existing literature on workload estimation and introduces a new long short-term memory (LSTM) model designed for workload prediction. The research relies on a dataset collected within the LUSAC laboratory and compares the performance of the proposed LSTM model with other approaches documented in the literature. The results demonstrate the effectiveness of the model in accurate workload prediction and estimation, addressing the needs of cloud users for the allocation of necessary data center resources. The evaluation employs various precision measures to highlight the capability of the developed model in workload estimation.