Distributed Data Deluge (D3): Efficient state management for virtualized Network Functions

Abstract

The advent of Network Function Virtualization (NFV) opens up new opportunities for scaling out network function (NF) instances to meet evolving demands. However, dynamic NF scaling in conjunction with the highly unbalanced Internet traffic can induce a high degree of processing load imbalance among the NF instances. In this respect, we present Distributed Data Deluge (D3) for the distribution of processing load among the virtualized NF instances. D3 uses forwarding and NF internal state distribution to achieve load balancing and further retain it upon NF instance scale in/out or NF state corruption. To this end, D3 leverages on Distributed Hash Tables (DHT) for state distribution. Since DHT cannot cope with the unbalanced Internet traffic, we augment DHT with redundancy and a flow handover mechanism. We further discuss techniques for detection and recovery from NF state corruption. Finally, we assess the load balancing efficiency of D3, using trace-driven simulations.