Multi-Controller Placement for Load Balancing in SDWAN

Abstract

The capability of flexible network management makes it available to incorporate softwaredefined networking (SDN) in the wide area network (WAN). Thereinto, multiple controllers are deployed in the software-defined wide area network (SDWAN) to tackle the performance bottleneck. Thus, how to effectively solve the load balancing problem in the control plane towards SDWAN under different constraints is worthy of further study. In order to achieve load balancing among the distributed controllers, we proposed a low-complexity controller placement algorithm, Simulated Annealing Partition-based K-Means (SAPKM), towards SDWAN. Meanwhile, two cost functions were proposed as the load balancing indices to assess the efficiency of the proposed algorithm from the perspective of topology structure and flow traffic distribution, respectively. Simulations were performed by using actual topologies on Topology Zoo with different sizes and different structures. Experimental results demonstrated the effectiveness of SAPKM in reducing both average load and load balancing indices with propagation latency and network reliability performance enhanced simultaneously. INDEX TERMS Software-defined wide area network (SDWAN), controller placement, load balancing, propagation latency, network reliability.

A new optimization technique to solve the latency aware controller placement problem in software defined networks

“…As you can see from Figure 5A, the gold color dots which represent the "$-method" is much lower than the others. 12,14 But, in the case of deploying 5 and 6 controllers in the OS3E network, the results of the CNPA 12 algorithm represented by the red dots are exactly the same as what we have found using the "$-method." In Figure 5B, we have shown our results for deploying controllers from 4 to 8 because RMM-FB and RMM-LM presented in Reference 11 was considered controller failure scenarios to improve the network resilience.…”

“…In Figure 5, we have shown the results of L w and compared it with other existing algorithms. [11][12][13][14][15] The authors of those references used different networks to show their results in their manuscript. The authors did not compare their algorithms for all these three networks (ie, OS3E, AT&T, and GEANT).…”

“…After that, we find the maximum number of incorporated columns (highlighted using coral color) residing only these red-colored rows. As we can see from Figure 1, we have two red-colored rows (i.e., 20,21) and after going through all the column numbers, we can see that there is no non-$ value in the column numbers (1)(2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14)(15)17) and row numbers (20,21). But, rest of the column numbers are our incorporated columns [i.e., {q} = {16, 18-25}] having two non-$ data which is highlighted using coral color in Figure 1.…”

“…To analyze the performance of our "$-method" with other existing algorithms, [11][12][13][14][15] we have used the following performance matrices such as worst-case SC latency (L w ), critical number of controller required to deploy in a network, and coefficient of variation (CV).…”

A comprehensive study of different objectives and solutions of controller placement problem in software‐defined networks

“…To achieve a load balance control plane, a multi‐controller placement algorithm for SDN has been proposed by the authors 70 that further enhanced the network reliability and propagation latency. The proposed CPP was formulated as MILP to minimize the average controller load as well as other metrics such as propagation latency and reliability.…”