A comparative study of failover schemes for IaaS recovery

Uninterrupted functioning of cloud infrastructure is critical to many everyday personal and professional activities. Preventing natural disasters from affecting cloud services is, therefore, important. Failover mechanism is an important component of any disaster recovery scheme. We compare three commonly used failover schemes analytically as well as empirically and present a Markov chain based generic model of disaster recovery. We implement disaster recovery approaches studied analytically in OpenStack and report results from several experimental settings. Analytical and empirical results suggest that having multi-level redundancy makes it more likely to recover quickly from different disaster scenarios.