Manifold learning in SLA violation detection and prediction for cloud-based system

Abstract

SLA is a contract between service providers and consumers, mandating specific numerical target values which the service needs to achieve. For service providers, detecting and preventing SLA violation becomes very important to enhance customer trust and avoid penalty charges. Therefore, it is necessary for providers to detect and forecast possible service violations. However, this is difficult to achieve when dealing with service violation in a cloud-based system due to multiple Quality of Service (QoS) parameters. In this work, manifold learning is used to reduce the high dimensionality problem arising from multiple QoS data into 1-D output of violation level (low, medium, high) data. From the transformed data, service violation will be detected as well as predicted based on violation level data. The violation level is obtained from aggregate value of each QoS weightage. Based on QoS data of 14 days, manifold learning is able to scale down 5 various parameters into a single parameter before detection and prediction process is performed. The prediction accuracy of Support Vector Regression as the time series analysis technique used is able to achieve 80%.