Assessment of Connectivity-Based Resilience to Attacks Against Multiple Nodes in SDNs

Abstract

In Software Defined Networks (SDNs), the control plane of a network is decoupled from its data plane. For scalability and robustness, the logically centralized control plane is implemented by physically placing different controllers throughout the network. The determination of the number and placement of controllers is known as the Controller Placement Problem (CPP). In the regular (i.e., failurefree) state, the control plane must guarantee a given maximum delay between every switch and its primary controller and a given maximum delay between every pair of controllers. In general, these delay bounds allow multiple solutions and, so, other goals can be used to determine the best CPP solution. In this paper, we assess the connectivity-based resilience to malicious attacks against multiple network nodes of the CPP solutions obtained with three different aims: the regular state delay optimization without any concern about attacks, the regular state delay optimization taking into consideration the worst-case attacks and the resilience optimization to attacks against multiple nodes. We assess the CPP solutions considering attacks of targeted nature (when the attacker has complete knowledge of the data plane) and attacks of nontargeted nature (i.e., random and epidemic attacks). We present computational results providing an analysis of the CPP solutions to the different types of attacks. The main conclusion is that the connectivity-based resilience between the different CPP solutions strongly depends on the network topology, the regular state delay bounds and the type of attacks. Finally, we provide insights on how SDN operators can consider the conducted assessment when deciding the controller placements in their networks.

Maximizing SDN resilience to node‐targeted attacks through joint optimization of the primary and backup controllers placements

“…Most of the works dealing with the problem of SDN network resilience against node‐targeted attacks assume that the attacker has full knowledge of the topology of the data plane network, but not about the actual location of the controllers, and knowingly selects the attacked nodes (and in this way constructing a node‐targeted attack) so as to achieve the most damaging effect on the network graph connectivity; such attacks are also called topological attacks or critical targeted attacks, see [4]. Recently, [23] presented an assessment of the resilience of long‐distance SDN networks to node‐targeted attacks. The paper compares the resilience of different controllers placement solutions to node‐targeted attacks and to node attacks of non‐targeted nature (in particular, random attacks), and one of the findings of this research is that node‐targeted attacks are much more damaging.…”

“…The first network ( cost266 ), consisting of $$$ V=37 $$$ node locations and $$$ E=57 $$$ links, is depicted in Figure 5 (and, for that matter, also in Figures 1 and 2); it is described in SNDlib [14] at http://sndlib.zib.de. The second network ( coronet conus ), composed of $$$ V=75 $$$ node locations and $$$ E=99 $$$ links, is shown in Figure 6; it is described for example in [25], [23].…”

Optimizing primary and backup SDN controllers' placement resilient to node-targeted attacks

“…The time interval until the complete network state view is reached is, in the worst case, the longest delay among all pairs of controllers, and hence the maximum controller-controller (CC) delay must also be bounded. Finally, the controller placement must be resilient to network disruptions, such as natural disasters [2], multiple link failures [3] or node-targeted attacks [4]. To improve the SDN resilience to these kinds of disruptions, in general additional controllers are required.…”