Optimizing virtual backup allocation for middleboxes

In enterprise networks, network functions such as address translation, firewall and deep packet inspection are often implemented in middleboxes. Those can suffer from temporary unavailability due to misconfiguration or software and hardware malfunction. Traditionally, middlebox survivability is achieved by an expensive active-standby deployment where each middlebox has a backup instance, which is activated in case of a failure. Network Function Virtualization (NFV) is a novel networking paradigm allowing flexible, scalable and inexpensive implementation of network services. In this work we suggest a novel approach for planning and deploying backup schemes for network functions that guarantee high levels of survivability with significant reduction in resource consumption. In the suggested backup scheme we take advantage of the flexibility and resource-sharing abilities of the NFV paradigm in order to maintain only a few backup servers, where each can serve one of multiple functions when corresponding middleboxes are unavailable. We describe different goals that network designers can take into account when determining which functions to implement in each of the backup servers. We rely on a graph theoretical model to find properties of efficient assignments and to develop algorithms that can find them. Extensive experiments show, for example, that under realistic function failure probabilities, and reasonable capacity limitations, one can obtain 99.9% survival probability with half the number of servers, compared to standard techniques.