Assessing Profit of Prediction for SDN controllers load balancing

Abstract

Software-defined networking (SDN) provides programmable control and centralized management in data centers, making it a popular architecture. The large scale of networks has required to propose the geographical distribution of logically centralized control plane to achieve scalability and reliability. For solving the load imbalance among multiple controllers associated with the statically configured control plane, a switch migration mechanism is proposed to admit dynamic load balancing. Many studies have been carried out for solving the control plane load balancing problem based on the switch migration mechanism. However, previous studies focus on migrating the switches when the controllers are overloaded, thereby, wasting time in the switch migration phase and resulting in high latency. To address these problems, we propose the Assessing Profit Of Prediction (APOP) scheme, a load-balancing strategy in the multiple-controllers control plane based on the overloaded state prediction and profit assessment. We introduce Taylor’s formula to predict the flow change in the network and assess the profit of migrating switches in advance, in order to decrease the migration time and minimize the harmful effects during the migration phase. The result of simulation experiments shows that our scheme performs effectively in reducing the migration cost in control plane load balancing.