Evaluating allocation heuristics for an efficient virtual Network Function chaining

Abstract

Enterprise networks widely deploy middleboxes to apply load-balancing techniques, to enforce policy compliance, and to improve security. Middlebox platforms, however, are closed systems and expensive. In turn, Network Function Virtualization (NFV) allows to deploy packet-processing middleboxes as virtual network functions, and to decouple the function from the physical realization. In this paper, we address the challenge of efficiently chaining virtual network functions. We propose and compare four heuristics for allocating virtual network functions over a network topology. Our proposal focuses on a greedy algorithm that allocates on demand a sequence of virtual network functions. We compare our four heuristics: (i) minimum introduced latency between source and destination nodes; (ii) minimum resource usage on the network nodes; (iii) the most central nodes first; and (iv) weighted decision between minimum latency and resource usage. We simulate our proposal over a real network topology, and the results show that we allocate 53% more requests when using the resource usage heuristic, and we reduce into 52% the average delay when using the latency heuristic.