Data-Center Architecture Impacts on Virtualized Network Functions Service Chain Embedding with High Availability Requirements

Network Functions Virtualization (NFV) is a recent networking trend gaining a lot of attention from telecom operators and vendors. It promises to virtualize entire classes of network node functions within a data-center and to deliver network services in the form of Virtualized Network Function (VNF) service chains using commercial off-the-shelf hardware and IT virtualization technologies. However, availability gets an important issue when purpose-built telecom hardware designed for the "fives nines" standard via built-in failure protection and recovery mechanisms is replaced by the off-the shelf hardware. With commercial off-the-shelf data-center hardware, failure probabilities could be higher than in traditional physical network infrastructure. Thus with NFV, infrastructure availability has to be considered all the way from the physical right up to the hypervisor layer and resilience mechanisms need to be built into the software and service provisioning design. In this work, we model different backup strategies for VNF service chains and provide algorithms for their resilient embedding in the data-center. Further, we answer the question which data-center topologies offer the best cost-per-throughput relation for a given resilience/availability for VNF service chains.