Quick Reboot-Based Recovery for Commodity Operating Systems in Virtualized Server Consolidation

Kernel-level failures have a considerable amount of impact on the overall availability of software systems. If a kernel fails, all the applications running on it also fail even if they are highly available. Reboot-based recovery is a simple but powerful way to recover from and avoid kernel failures. From the authors' practical experience, they know that OS reboots successfully recover from kernel failures even if their root causes are unknown. Unfortunately, rebooting a commodity OS involves significant downtime, which is unacceptable in commercial services. This paper presents FlashReboot, a lightweight reboot technique that shortens downtime involved in reboot-based recovery.