POWER MANAGEMENT APPROACHES FOR VIRTUALIZED SERVER CLUSTER ENVIRONMENT

Abstract

Cloud computing paradigm provides Information Technology (IT) newlineinfrastructure, resources and services. Wider adoption of virtualization newlinetechnologies and cloud models has paved the way for easy access, ondemand newlineresource capacity availability, better operational controls, pay-per-use newlineetc. newlineEnergy consumption is a significant operational expense cost driver to newlinea cloud service provider. In Cloud computing, it is crucial to schedule tasks in newlinea manner that exploits the heterogeneity of the resources and applications to newlineoptimize systems performance. For e.g., it is important that right server is newlineselected and also the right quantum of such servers is maintained in newlineoperational state to process workload or service requests. newlineHolistic datacenter energy minimization operation should also consider newlineusage pattern interactions between computing and cooling sources. Decisions newlinelike workload type, server configuration, load, utilization etc., contributes to newlinepower consumption and influences datacenter s thermal profile which in-turn newlineimpacts the energy required to control server cluster component temperature newlinewithin operational thresholds. newlineThis thesis proposes power management approaches to improve newlineenergy efficiency of a heterogeneous virtualized server cluster system, newlineespecially accounting for server processor sleep states, sleep state power and newlinetransition latency overheads, and thermal operability at datacenter level. newline