Abstract

Abstract. Cloud services are becoming centralized at several geo-replicated datacentres. These services replicate data within a single dat-acentre to tolerate isolated failures. Unfortunately, the effects of a dis-aster cannot be avoided, as existing approaches migrate a copy of data to backup datacentres only after data have been stored at a primary datacentre. Upon disaster, all data not yet migrated can be lost. In this paper, we propose and implement SDN-KVS, a disaster-tolerant key-value store, which provides strong disaster resilience by replicating data before storing. To this end, SDN-KVS features a novel commu-nication primitive, SDN-cast, that leverages Software Defined Network (SDN) in two ways: it offers an SDN-multicast primitive to replicate critical update request flows and an SDN-anycast primitive to redirect request flows to the closest available datacentre. Our performance evalu-ation indicates that SDN-KVS ensures no data loss and that traffic gets redirected across long distance key-value store replicas within 30 seconds after a datacentre outage. 1