Dynamic Reconfiguration for Service Function Chain Orchestration in Next Generation Networks

The unprecedented dynamics of network resources and service requirements for the next generation networks bring great challenges in Service Function Chain (SFC) orchestration. Most existing work has only focused on deterministic scenarios, where the resource capacities and service requests are assumed to be static and known in advance, and thus treated as a one-shot optimization problem. In this paper, the SFC orchestration problem with dynamic network resources and bursty service arrivals is studied. To cope with the challenges of complicated decision space and different time scales of network dynamics, a Two-Stage Dynamic Reconfiguration (TSDR) approach is proposed. TSDR solves the problem by dividing it into two stages: first to determine the SFC number by a reinforcement learning method and then to determine their embedding positions and the routing paths by a Heuristic SFC Migration (HSM) algorithm. By exploiting the carefully designed weights and the depth-first search, the HSM algorithm is shown to achieve near-optimal performance and be adaptive to dynamic resource capacities. Numerical results show that the proposed TSDR approach is robust to the burstiness of service arrivals while achieving a desired tradeoff between the end-to-end service delay and deployment costs.