Management of high availability services using virtualization

Abstract

This thesis examines the use of virtualization in management of high availability services using open source tools. The services are hosted in virtual machines, which can be seamlessly migrated between the physical nodes in the cluster automatically by high availability software. Currently there are no complete open source solutions that provide migration of virtual machines as a method for repair. The work is based on the high availability software Heartbeat. In this work, an add-on to Heartbeat is developed, allowing Heartbeat to be able to seamlessly migrate the virtual machines between the physical nodes, when shut down gracefully. This add-on is tested in a proof of concept cluster, where Heartbeat runs Xen virtual machines with high availability. The impact of migration has been measured for both TCP and UDP services, both numerically and heuristically. The outages caused by graceful failures (e.g. rebooting) are measured to be around 1/4 seconds. Practical tests are also performed. The impression is that the outages are not noticed by the users of latency critical services as game servers or streaming audio servers.