An integrated framework of multi-controller management scheme for highly reliable Software Defined Networks

• Main Authors: Yao-ying Tzeng, 曾耀瑩
• Other Authors: Chung-An Shen
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/e2x65u

碩士 === 國立臺灣科技大學 === 電子工程系 === 107 === Since the scale of the network has increased, the scalability challenge of deploying the multi-controller in Software-defined Networking becomes more important. In addition, to stably provide the function of the control plane, improving the reliability of the control plane is also an important consideration in multi-controller deployment. In order to simultaneously consider the scalability and reliability of multi-controller deployment, this paper proposed an integrated multi-controller management scheme for highly reliable Software-defined Networks to deploy multi-controller. The management scheme included two stages: Network Planning and Runtime Maintain. The Network Planning section contained three parts to find the initial deployment of control plane. (1) Controller Quantity: estimate the required number of controllers with synchronization cost. (2) Controller Placement: using the Maximum Neighbors Controller Placement algorithm (MNCP) to find the situation which has lower isolation probability. (3) Switch Assignment: planning the higher device connectivity management of each controller with Load Based Hop Count Assignment algorithm (LHCA). And in the Runtime Maintain section, it used the State Behavior Tree, which executes the corresponding behavior according to the current state of controllers (e.g. Controller loading percentage, overloading or recovery from the failure), and lower the probability of controller overload. The results of proposed framework show reducing up to 89% isolation probability and increase by up to 13% device connectivity with the prior work in the Network Planning part. And in the Runtime Maintain, all the controllers delayed the overload time and keep at least one controller alive while the load of network equal to the whole controller capacity.