Skip to Main Content
This paper discusses a novel data center architecture based on load-balanced forwarding with in-packet Bloom filters enabled by two support services that distribute the directory and topology state of OpenFlow controller applications. By deploying an army of Rack Managers acting as OpenFlow controllers, the proposed architecture promises scalability, performance and fault-tolerance. We conjecture that packet forwarding itself may become a cloud internal service implemented by leveraging cloud application best practices such as low-latency, highly-available, key-value storage systems. Moreover, we contribute to demystifying the argument that the centralized controller model of OpenFlow networks is prone to a single point of failure and show that direct network controllers can be physically distributed, yielding a “sweet spot” between fully distributed and centralized networking paradigms.