Improving the Responsiveness of Replicated Virtualized Services in Case of Overloaded Replicas Connectivity

One approach to provide high availability for services is to encapsulate the service in a virtual machine and replicate, in checkpoints, the entire machine. The replication protocol can overload the network link between the replicas, fact that will increase the latencies experienced by the clients of the service. We implemented a context-aware replication protocol that adapts the length of a checkpoint phase in order to get low values for the latencies experienced by the service's clients. The length of a checkpoint phase is obtained by using predicted parameters that describe the service behavior. The results show us that the latencies experienced by the service's clients were reduced, especially in case the network link between replicas was overloaded. Our replication protocol extends Remus replication protocol, which is provided with Xen Hyper visor, by improving its asynchronous strategy and by taking into account the context of the replicated service.