Modified Floyd-Warshall Algorithm for Equal Cost Multipath in Software-Defined Data Center

Load balancing in data centers have been a common practice in the last couple of decades. This has been done statically in traditional networks with little or no feedback information from the underlying network state. With the current large cloud data centers and continuous changing traffic patterns, the drive for more interactive and dynamic solution to reduce the latency and improve the network resources utilization has brought about the Software Defined Networking Paradigm. However, some load balancing solutions have been proposed, utilizing OpenFlow without a well-defined algorithm to reduce the path computation complexity as requests arrive on the network. This paper proposes a path load-balancing algorithm which utilizes a modified Floyd-Warshall All-Pairs Shortest Paths algorithm to compute and store equal cost paths information, utilizes the stored information for path selection, and maintain a real-time updates of those paths. Our evaluation demonstrates that the proposed algorithm performs better than the Global First Fit algorithm as we have significantly reduced the time to service request as a result of eliminating the recursive path computation for every client request.