High-Availability Service Chain Realization Theory

We propose a novel approach to realize service chains with specific availability and latency requirements by dimensioning a NFV infrastructure (NFVI). Our contribution leads to a theory that is based on extending SDN capacity computation to the domain of NFV. We do so by expanding upon the requirements of availability of service chains (SCs), projecting these onto a NFVI. A key contribution of our theoretical model is the inclusion of delay bounds while traversing a SC. The combination of availability, delay bounds and computing SC granularity is achieved through modeling SCs as directed acyclic graphs in a NFVI, and acted upon using network calculus, eventually leading to a SC realization theorem. This SC realization theorem answers a question as to how many SCs of a particular type and granularity can be provisioned on a particular DC. A second fast approach towards computing SCs realization on a NFVI is also presented. The theorems are verified numerically for DCs of size upto 12,228 servers (nodes). Results for scalability and latency are presented.