Two-phase load balancing of In-Memory Key-Value Storages using Network Functions Virtualization (NFV)

Abstract Social networks and other clouding applications should require fast responses from datacenter's infrastructure. One of the techniques that have been widely used for achieving such requirement is the employment of In-Memory Key-Value Storage (IMKVS) as caching mechanisms in order to improve overall user experience. Memcached and Redis are applications that use IMKVS approach. Commonly IMKVS systems use Consistent Hashing to decide where to store an object, which may cause network load imbalance due to its simplistic approach. Furthermore, these systems work only at the application layer, so network conditions are not considered to distribute user's accesses. This paper proposes a new cache architecture based on a two-phase load balancing to improve IMKVS performance, which has adopted a Network Function Virtualization (NFV) architecture to manage the load balancing mechanism. The proposal was evaluated in the Mininet and shows an improvement of 23% on the load of the caching servers and 5% on the load of the network compared to Consistent Hashing, which results in better resource usage and better user experience.