Quantifying the performance impact of overbooking virtualized resources

Cloud services are compelling to customers from many perspectives, but the immense benefits come at a price. Along with reliability and security, performance is the Achilles heel of Cloud services. In order to make Cloud services profitable, the Cloud service provider obviously must overbook physical resources. But at what level of resource overbooking (OB) does performance begin to suffer? The answer, of course, depends on many factors. In this paper, we present a simple analytic model to quantify the performance impact of overbooking virtualized resources as a function of the relevant environment and usage parameters. We then validate those analytic modeling results against simulation, lab tests, and field data. Finally, we propose a simple means for measuring the model parameters in the field, in order to use the analytic model to determine allowable OB factors (engineering rules) while still meeting performance constraints and service-level agreements.