Research On The Mechanisms Of Virtual Machine Replication Oriented High Availability Services

Virtualization technology has been widely used in cloud computing,data center and cyber physical systems.However,with the system scale expanding unceasingly and business lines growing continuously,how to ensure the system running stably and efficiently,how to reduce the impact of unplanned downtime,for providing a more stable and highly available basic platform has been the top priority of computer system development.Virtual Machine(VM)replication technology which periodical or using event channel tapping mechanism replicating the running VM state on primary node to backup VM on secondary node is a well known technique for supporting HA service.Checkpoint-recovery based VM replication taking advantage of redundant physical resources to provide non-stop service.However existing VM replicaion protocol aiming for guarantee state consistency by importing network and disk buffers has brought about greatly reducing availability of VM replication,especially in case the network link between replicas was overloaded.Therefore the exisiting VM replication protocol cannot satisfy the technical requirements of current environment.How to remedy the deficiencies and improve the availability of service has become a pressing problem.This paper briefly describes the background and develop process about VM replication,explains the key technology of VM replication,focuses on representative VM hot-standby system Remus.The paper proposed an adapative VM repliciton system architecture based on Remus.According to fine-grained partition about replication protocol,adaptive VM replication protocol oriented Xen is presented on purpose with impoving service availability.This mechanism adapts to the running context,consisting of the behavior of the service.The length of each replication phase is determined dynamically in order to improve as much as possible the latencies experienced by the clients of the service.At the meantime the protocol taking availability factor into account.Besides,the paper proposes a method to maintain consistency based on BDwrite to ensure the VM data and state consistency.The experiment result proves that the adaptive mechanism implementing service latencies estimate precisely.The latencies experienced by the clients are significantly reduced,especially in case the network link between replicas is overloaded.