Experiences with BGP in large Scale Data Centers
Abstract
Large scale data centers support heterogeneous cloud workloads that include search, PAAS/SAAS, and Hadoop-like compute and storage.
The datacenter networks deployed in these data centers must scale to hundreds of thousands of server nodes while optimizing the cost per megabit of providing the required bisection bandwidth.
To meet these requirements, we deploy a Layer-Three-to-the-TOR CLOS/Fat Tree topology with multivendor switching elements and high link density.
We will discuss our use of BGP as the only routing protocol in this network. BGP provides scalability, stability, and reliability, and is well supported by multiple vendors. Using a BGP speaker, a path computation engine, and a controller hosted in a virtualized server environments, we will show how we can leverage BGP’s standard feature set to implement a hybrid SDN solution.
Presenter
Edet Nkposong (Microsoft)
Tim LaBerge (Microsoft)
Naoki Kitajima (Microsoft)
Slides