An Efficient Route Management Framework for Load Balance and Overhead Reduction in SDN-Based Data Center Networks

Abstract

A data center network (DCN) is composed of many servers interconnected by well-organized switches, which involves massive amount of data transmissions to provide cloud services. It is critical to manage packet routes in DCNs to avoid congestion on some links. The software-defined networking (SDN) technique supports auto-configuration of switches by a central controller, and gives an easy way to manage traffic flows. In the paper, we exploit SDN to improve performance of fat-tree DCNs, and propose a low-cost, load-balanced route management (L2RM) framework. L2RM keeps monitoring network traffics and computes a load-deviation parameter to check if some links of switches are burden with heavy loads. Then, an adaptive route modification mechanism is triggered if necessary, which considers the size limitation of flow tables and uses group tables in OpenFlow to distribute flows among different links to balance their loads. Besides, L2RM uses a dynamic information polling mechanism to query switches about their statuses, so as to reduce the message overhead of the controller. Through Mininet simulations, we show that our L2RM framework can better increase link utilization, alleviate table overflow, and reduce message cost, as comparing with other SDN-based approaches for fat-tree DCNs.