A Proactive Mechanism To Improve Workload Prediction For Cloud Services Using Machine Learning

Gursale, Sumedh (2020) A Proactive Mechanism To Improve Workload Prediction For Cloud Services Using Machine Learning. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration manual) Download (1MB) | Preview

Abstract

Service elasticity is an important enabler of cloud computing. It is nothing but the ability to adapt the system to inconsistent changes in workload by dynamic provisioning and de-provisioning services so that the unused resources suit the current demand at all times. This requirement of users is fulfilled by almost all cloud service providers. However, one of the major challenges cloud service providers face is the efficient resource allocation as per the demand changes and maintaining the quality of services (Qos) as per service level agreement (SLA’s). The service providers don’t fulfill the SLA and reason being is an unavailability of demand for workload that could lead to downtime because of heavy traffic across the network and to avoid this all cloud service providers offers standard solution of over provisioning to support peak load and guarantee of maintaining QoS over lifetime of operation. This causes in resource consumption and is not cost-effective as the machine stays idle most of the time and contribute to greater power utilization. This paper focuses on implementing a hybrid prediction approach based on machine learning techniques. In first stage, the focus is to break down or split time series data input signal into two parts. Secondly to predict low frequency components, Support Vector Regression (SVR) is used on first part. Second part of time series is more likely noise and has high frequency, so for the prediction Artificial Neural Network (ANN) is used. Lastly, an inverse wavelet transformation is implemented to reconstruct these samples to original signal from two multi-scale predictions in order to achieve accurate workload prediction. Based on the overall results, the proposed approach has a relatively better predictability compared with competitive approach. The results are evaluated on two models and proposed model (A hybrid SVR + ANN) has outperformed the other model.

Item Type:	Thesis (Masters)
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Cloud computing
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Dan English
Date Deposited:	28 Jan 2021 14:09
Last Modified:	28 Jan 2021 14:09
URI:	https://norma.ncirl.ie/id/eprint/4536

Actions (login required)

View Item