Paratus: Instantaneous Failover via Virtual Machine Replication

Business continuity is valuable to IT management in both small and large organizations. As faults, errors, and even disasters perfectly exist, instantaneous service fail over must be furnished to computer systems in order to preserve business continuity. However, current solutions concentrate either on volatile memory and CPU context or on persistent disk storage, instantaneous fail over can hardly be achieved as expected. To remedy this, we design a novel framework that encapsulates computer system in a virtual machine, and henceforth frequently replicates its state to a backup physical host. In case the primary virtual machine crashes, the virtual machine replica at the backup host can instantaneously take over the failed virtual machine to service unfinished and new tasks without user's perception. We design and implement a prototype system, Paratus, to demonstrate this argument. Experimental results show that seamless service recovery is accomplished within seconds at the cost of no consistency check, restart, or reconfiguration required. Meanwhile performance degradation is hardly visible to users.