On Ensuring Full Yet Cost-Efficient Survivability of Service Function Chains in NFV Environments

The emergence of Network Function Virtualization enables the deployment of network services in the form of service function chains. In this context, one of the key challenges is to ensure the survivability of these chains in face of single or multiple simultaneous physical node failures. In this paper, we address this challenge and propose solutions to guarantee the survivability of service chains by ensuring that there are enough backups ready to take over when failures occur. Specifically, we put forward a Survivability Management Framework that predicts traffic demand in service function chains and provision enough backups for network functions with minimal costs. To this end, we leverage the AutoRegressive Integrated Moving Average (ARIMA) model to predict future demand. We mathematically model the service chain survivability problem as an integer linear program that determines the minimal number of shared backups and their optimal location in the infrastructure such that backup operational costs are minimized. We also devise two greedy algorithms to deal with the problem in large-scale scenarios. We show, through several simulations, the performance and efficiency of the proposed solutions in different scenarios. We also show that demand prediction could help to avoid unnecessary provisioning of backups, and thereby reduce their operational costs.