virtual machine replay update: improved implementation for modern hardware architecture

This paper describes a successive and updated work of Revirt project which presents a virtual machine replay framework on Xen hypervisor. As both the commodity hardware and Xen hypervisor have been changed significantly since the first publication of Revirt, the initial implementation does not meet the needs of modern architecture any more. This paper presents an improved implementation of virtual machine execution replay system called CASMotion. CASMotion has three contributions. First, CASMotion uses the performance monitor of Intel Core2 processor to construct time point of recorded events, which makes the event record more complete and precise. Second, CASMotion can fully support multi-core hardware platform which is prevalent today. Third, CASMotion is developed with more general architecture design, which makes it deployable on upstream Xen hypervisor and Dom0. Our experiments under a varity of workloads shows CASMotion has low performance impact on monitored DomU. The growth of record log is also in acceptable range. Index Terms-Revirt, execution replay, determinism, virtual machines, Xen. © 2012 IEEE.IEEE Reliability SocietyThis paper describes a successive and updated work of Revirt project which presents a virtual machine replay framework on Xen hypervisor. As both the commodity hardware and Xen hypervisor have been changed significantly since the first publication of Revirt, the initial implementation does not meet the needs of modern architecture any more. This paper presents an improved implementation of virtual machine execution replay system called CASMotion. CASMotion has three contributions. First, CASMotion uses the performance monitor of Intel Core2 processor to construct time point of recorded events, which makes the event record more complete and precise. Second, CASMotion can fully support multi-core hardware platform which is prevalent today. Third, CASMotion is developed with more general architecture design, which makes it deployable on upstream Xen hypervisor and Dom0. Our experiments under a varity of workloads shows CASMotion has low performance impact on monitored DomU. The growth of record log is also in acceptable range. Index Terms-Revirt, execution replay, determinism, virtual machines, Xen. © 2012 IEEE