B-Alarm: An Entropy Based Burst Traffic Prediction Approach for Ethernet Switches in Data Centers

Abstract

With the rapid growth of data centers, research on Ethernet switches in data centers have become a much more critical issue. Since Ethernet switches are essential to the performance of data center networks and can consume around 15-20% of the total energy in data centers. This makes monitoring and prediction of traffic patterns very significant. Most previous studies on traffic patterns in data center switches employ passive monitoring and cannot effectively follow sudden traffic bursts. However, in data center networks, the traffic flow can change a lot very rapidly. An active predicting method for traffic bursts in data center switches is the key to improve the performance of these switches. Therefore, to overcome the limitations of previous works and solve this problem, we introduce a novel entropy based prediction approach named "B-alarm" to provide an active prediction mechanism of incoming traffic bursts. Simulation experiments are conducted using ns-2 platform. The results show that our burst prediction approach can achieve quick and accurate burst traffic prediction in data center switches with an accuracy of at least 85%.