Provably Efficient Algorithms for Traffic-sensitive SFC Placement and Flow Routing
Yingling Mao, Xiaojun Shang, Yuanyuan Yang
Abstract
Network Function Virtualization (NFV) has the potential of cost-efficiency, manage-convenience, and flexibility but meanwhile poses challenges for the service function chain (SFC) deployment problem, which is NP-hard. It is so complicated that existing work conspicuously neglects the flow changes along the chains and only gives heuristic algorithms without a performance guarantee. In this paper, we fill this gap by formulating a traffic-sensitive online joint SFC placement and flow routing (TO-JPR) model and proposing a novel two-stage scheme to solve it. Moreover, we design a dynamic segmental packing (DSP) algorithm for the first stage, which not only maintains the minimal traffic burden for the network but also achieves an approximation ratio of 2 on the resource cost. Such a two-stage scheme and DSP can pave the way for efficiently solving TO-JPR. For example, simply applying the nearest neighbor (NN) algorithm for the second stage can guarantee a global approximation ratio of O(log(M)) on the network latency, where M is the number of servers. More future work can be done based on our scheme to get better performance on the network latency. Finally, we perform extensive simulations to demonstrate the outstanding performance of DSP+NN compared with the optimal solutions and benchmarks.