Prolog to "Managing Performance Overhead of Virtual Machines in Cloud Computing: A Survey, State of the Art, and Future Directions"
Jim Esch · Proceedings of the IEEE · 2014
Due to its scalability and on-demand resource flexibility, cloud computing has emerged as an economical means for pay-as-you-go computing. Infrastructure-as-a-Service (IaaS) cloud platforms host large and successful company services; however, virtual machine (VM) performance is unpredictable, which can noticeably impact a tenant's service-level agreement. Resource contention between VMs is the main culprit; it leads to degradation and variation, also known as “performance overhead.” Overhead can be so severe that it will hinder users from moving critical business services into the cloud. Overhead manifests alternately as increased execution time, decreased network throughput, and uneven application performance over a period of time. Performance overhead has been studied under three scenarios: the single-server virtualization, the single large datacenter, and multiple geodistributed datacenters.