Developing Machine Learning and Deep Learning Models for Host Overload Detection in Cloud Data Center
Manzano S. Ricardo, Nishith Goel, Marzia Zaman, Rohit Joshi, Mustafa Daraghmeh, Anjali Agarwal · 2021 IEEE 12th Annual Information Technology, Electronics and Mobile Communication Conference (IEMCON) · 2021
Cloud service providers need to deal with many challenging problems to provide services such as guaranteeing Service Level Agreement (SLA) with clients and minimizing energy consumption of the data centers. Cloud resource management tackles these problems using different techniques such as consolidation. In consolidation, the migration of VMs depends on host overloading which evaluates the state of a host (overloaded or under loaded) before migrating a VM. Forecasting the state of a host accurately and in time is crucial to guarantee SLA with clients and reduce energy consumption in a datacenter. CPU utilization is used in the present work to determine the state of the host. In this paper, we propose a methodology to evaluate how eight different algorithms including machine and deep learning, which forecast the servers' CPU utilization, impact the migration of VMs affecting SLA and energy consumption in datacenters. The outcomes enable decision-makers to take better decisions which result in the reduction of energy consumption and improvement in SLA with clients. The performances of the models are evaluated with random and real users' workloads. We propose a methodology to develop, evaluate and implement host overload detection models in cloud data center based on machine learning (ML) and deep learning (DL) techniques. We validate the methodology and demonstrate that Long Short-Term Memory (LSTM) and Simple Moving Average (SMA) provide excellent results with on average over 2.5 times reduction of Energy and SLA Violation (ESV) metric when compared to other algorithms. Also, LSTM is found to be the most robust solution to predict CPU utilization using unseen data.