CO-BPG: A Centralized Optimizer for Routing in BGP-Based Data Center Networks

Load-balancing routing is an important and challenging issue in data center networks. Traditional internet routing protocols like OSPF and BGP cannot provide satisfiable network throughput because they are traffic-unaware and cannot utilize the rich connections and topology regularity in data center networks. New routing approaches proposed for data center networks can achieve good load balance, but cannot deploy widely because they often introduced in special switch/host hardware or software. In this paper, we propose a centralized routing optimizer, CO-BGP, for data center networks using BGP as a base routing protocol. It provides routing optimization with minimal deployment cost with off-the-shelf data center switches and servers. A routing optimization algorithm is proposed to optimize the routing performance by taking into account of network traffic load. The algorithm uses Bayes learning to cope with the dynamic of network traffic. Experiment results have shown that the proposed algorithm has better performance than the well-known equal cost multi-path (ECMP) approach.