Achieve High Availability Of Network System Through Active-backup Virtual Machine

With the development of the Internet, network servers are ubiquitousin various areas. To achieve high availability is quite significant fornetwork server since service failures will cause large losses. Virtualmachine (VM) replication (replicating the state of a primary VM runningon a primary node to a backup VM running on a secondary node) is wellknown technique for providing non-stop service. Unfortunately, existingVM replication approaches suffer from excessive replication overhead.In this paper, we propose Shadow VM, a generic and highly efficientnon-stop service solution, based on VM replication. Shadow VM forwardsrequest from client to both primary VM and backup VM, and monitors theoutput responses of the primary and backup VMs. Shadow VM considersthe backup VM as a valid replica of the primary VM, as long as networkresponses generated by the backup VM match that of the primary. Theprimary VM state is propagated to the backup VM if outputs from thebackup VM and primary VM servers no longer match.By studying the TCP/IP stack, we eliminate several non-deterministicfactors of TCP/IP stack. Therefore, the output similarity between primaryVM and backup VM is improved. We also dig into the process of VMreplication and reduced its overhead. The experimental result shows thatShadow VM can deliver an average30%of native performance for Weband FTP server benchmark workloads. Furthermore, Shadow VMintroduces little network latency and can failover within a few hundred ofmilliseconds.