A Workload, Performance and Resource Usage

Abstract

This paper proposes a canonical correlation analysis (CCA) based workload-performance-resource (WPR) model which can capture and compare the complex many-to-many workload, performance and resource consumption relationship of an application running in physical and in virtual machines. The model can also establish complex relationships of the usage variables of four potentially interrelating resources (CPU, memory, disk I/O and network I/O) used by the application. The model is intended to be used in planning application resource requirements prior to cloud migration. Experimental results show that the WPR model can model and capture the complex resource consumption behavior of an application and the system modules that perform operations on its behalf, as well as the intricate correlation between the four types of resources, and gives good prediction performance.