Reliability in cloud computing: Downtime predictions for virtual servers

Reliability in cloud computing, i.e. failure resilience and availability, is a critically important and timely issue. Google, Amazon and other cloud computing service providers have all experienced severe service outages in the recent past. Having poor reliability in the cloud not only affects its existing users and applications, but also deters new users and applications from using cloud services, resulting in millions of dollars of losses in revenue and productivity. In this work, we address a fundamental question faced by cloud computing service providers pertaining to reliability, specifically, how to predict the downtime of provisioned server infrastructure for a given finite duration considering concurrent machine failures. We design a set of algorithms using an analytical and an empirical approach based on the limiting behavior of the birth-death process and sample path analysis to address the question, and provide a roadmap for the empirical analysis to be accomplished in future.