Distributed predictive performance anomaly detection for virtualised platforms. (2018)

• Record Type: Journal Article
• Authors: Jehangiri, Ali Imran; Yahyapour, Ramin; Yaqub, Edwin; Wieder, Philipp
• Is Part Of: International journal of high performance computing and networking. Volume 11:Number 4(2018)
• Page Start: 279
• Page End: 290
• Publication Date: 2018
• Subjects: cloud -- performance prediction -- monitoring -- quality of service -- QoS -- analytics -- distributed time-series database; High performance computing -- Periodicals; Computer networks -- Periodicals; High performance computing; Periodicals; 004.05
• Languages: English
• ISSNs: 1740-0562

Abstract

Predicting subsequent values of quality of service (QoS) properties is a key component of autonomic solutions. Predictions help in the management of cloud-based applications by preventing QoS breaches from happening. The huge amount of monitoring data generated by cloud platforms motivated the applicability of scalable data mining and machine learning techniques for predicting performance anomalies. Building prediction models individually for thousands of virtual machines (VMs) requires a robust generic methodology with minimal human intervention. In this work, we focus on these issues and present three main contributions. First, we compare several time series modelling approaches to evidence the predictive capabilities of these approaches. Second, we propose estimation-classification models that augment the predictive capabilities of machine learning classification methods (random forest, decision tree, and support vector machine) by combining them with time series analysis methods (AR, ARIMA and ETS). Third, we show how the data mining techniques in conjunction with Hadoop framework can be a useful, practical, and inexpensive method for predicting QoS attributes.