U-HAUL: Efficient State Migration in NFV

Libin Liu, Hong Xu, Zhixiong Niu, Peng Wang, Dongsu Han

Abstract: Network function virtualization (NFV) enables dynamic scaling of resources to middlebox deployment and management. In NFV, state migration is an important task because operators often need to shift traffic and its associated flow states across NF instances for load balancing. Existing state migration schemes, however, exhibit long delays and high controller overhead. This paper presents U-HAUL, an efficient state migration system that reduces the state migration overhead. U-HAUL takes advantage of the fact that most flows are short-lived mice flows, and in many cases their processing states will expire before the state migration finishes. Rather than blindly moving states of all the flows, U-HAUL keeps the states of active mice flows on the original NF instance, and only migrates elephant flow states. By reducing the number of flow states to be migrated, U-HAUL greatly reduces the migration delay and its performance penalty. Our evaluation shows that U-HAUL reduces the average migration time by up to 87% and the latency to mice flows by up to 94% compared to OpenNF.