Green Virtual Network Function Placement with Respect to Time-Varying Data and Potential Link Failures

Abstract: This thesis presents an integer programming model for energy efficient routing of network traffic and deployment of virtual network functions in a network. The optimization model provides certain protection against hardware failure, by finding both a primary and a backup path for each traffic demand. Furthermore, the model is capable of modelling multiple time periods, allowing the network to adapt to expected changes in network load. By applying Soyster's robustness principle, the model can also yield a network configuration that can handle network load that deviates from the expected load. Lastly, the model ensures an upper bound on the latency of service chains. As expected with integer programming models, this model is difficult to solve for anything but the smallest of networks. Therefore, a single time period and variable fixing heuristic is proposed to help speed up the solve time. Evaluation of the model and heuristic is done in the context of the Evolved Packet Core. The evaluation shows that the heuristic indeed improves the solve time significantly, but it also shows that there is room for improvement.