In Software-Defined Networks (SDN), one of the critical research challenges for large-scale deployment of SDN is controller placement problem (CPP). However, link failures are critical security issues in networks and greatly impact SDN’s robustness. The CPP for link failures is still challenging today. To this end, we study the CPP for multi-link failures (CPP-MLF). Firstly, we investigate the real link data of today’s networks to analyze the characteristics of link failures and formalize the link failure rate. Next, we formulate minimizing the number of controllers and worst-case delay as the optimization objectives of CPP-MLF and develop a heuristic algorithm based on improved NSGA-II to solve CPP-MLF efficiently. Specifically, we improve the non-dominated set and crowding distance for the NSGA-II by dynamic competition mechanism and even distribution operator, respectively, thereby quickly obtaining the Pareto-optimal solutions of CPP-MLF. Afterwards, we introduce controller load variance to evaluate them to make a satisfactory decision on controller placement. We conduct simulations with real network topologies, and the results show that the proposed heuristic algorithm performs well on the number of controllers, worst-case delay and robustness, while producing acceptable runtime overheads.

在软件定义网络 (SDN) 中，大规模部署 SDN 的关键研究挑战之一是控制器放置问题 (CPP)。然而，链路故障是网络中的关键安全问题，极大地影响了 SDN 的健壮性。用于链路故障的 CPP 今天仍然具有挑战性。为此，我们研究了多链路故障的 CPP (CPP-MLF)。首先，我们调查当今网络的真实链路数据，以分析链路故障的特征并形式化链路故障率。接下来，我们将控制器数量和最坏情况延迟的最小化作为 CPP-MLF 的优化目标，并开发基于改进的 NSGA-II 的启发式算法来有效地解决 CPP-MLF。具体来说，我们分别通过动态竞争机制和均匀分布算子改进 NSGA-II 的非支配集和拥挤距离，从而快速获得 CPP-MLF 的帕累托最优解。之后，我们引入控制器负载方差来评估它们，以便对控制器放置做出令人满意的决定。我们使用真实的网络拓扑进行模拟，结果表明所提出的启发式算法在控制器数量、最坏情况延迟和鲁棒性方面表现良好，同时产生可接受的运行时开销。