On the Usability of Virtualization Technologies for Application with Stringent Requirements

Virtualization technology has brought about key benefits - simplified management by decoupling the virtual machine from the physical hardware it runs on and allowing virtual machines to be dynamically migrated away from bottlenecked or faulty machines toward newer or better suited ones, better resource utilization by allowing consolidating multiple virtual machines onto fewer physical machines to save power, and strong isolation by ensuring that one faulty virtual machine cannot affect the performance or health of another. In this paper, we discuss a fourth benefit yielded by virtualization technology – generating virtual machines checkpoints (snapshots) so that they can be quickly restarted when they crash – and discuss the cost/benefits trade-offs between various checkpointing techniques. We evaluate taking periodic checkpoints on Market Matching, a real-world application with stringent availability and performance requirements, and show that although existing checkpointing mechanisms can be beneficial for applications with simple workloads, they are not practical for use with applications with stringent availability requirements. Finally, we discuss bottlenecks that slow down the process of taking snapshots in current hypervisor implementations.