Multi-Controller Placement for Load Balancing in SDWAN

Kongzhe Yang,Bangning Zhang,Bing Zhao,Daoxing Guo

doi:10.1109/access.2019.2953723

Kongzhe Yang, Bangning Zhang + Show 2 more

Open Access

https://doi.org/10.1109/access.2019.2953723

Copy DOI

Abstract

The capability of flexible network management makes it available to incorporate software-defined 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.

Highlights

Software-Defined Networking (SDN) is a new paradigm which departs the control plane from the data plane to further improve network performance and manage the whole network more efficiently and flexibly
By incorporating the centroid-based clustering algorithm [8], we propose the Simulated Annealing Partition-based K-Means (SAPKM) algorithm to solve the load balancing controller placement problem
We model the load balancing controller placement problem as a mixed-integer linear program (MILP) with performance metrics formulated mathematically

Summary

Introduction

Software-Defined Networking (SDN) is a new paradigm which departs the control plane from the data plane to further improve network performance and manage the whole network more efficiently and flexibly. Multiple physically distributed controllers composing the control plane offer a logically centralized vision of the network and manage forwarding devices in the data plane to forward packets and to execute the handover. E.g. the software-defined wide area network (SDWAN), need numerous dedicated controllers to meet performance metrics, including latency, reliability, etc. How to determine the number and the locations of the controllers with appropriate assignments for solving load balancing issue in SDWAN worth in-depth study. Heller et al [1] introduce the controller placement problem (CPP) for the first time. The authors study the impact of the number and locations of the controllers on both

Objectives

Results

Conclusion