A Scalable and Resilient Layer 2 Network With Ethernet Compatibility in Java

We present the architecture and protocols of ROME, a layer-2 network designed to be backwards-compatible with Ethernet and scalable to tens of thousands of switches and millions of end-hosts. Such large-scale networks are needed for emerging applications including data center networks, wide area networks, and metro Ethernet. ROME is based upon a recently developed greedy routing protocol, greedy distance vector (GDV). Protocol design innovations in ROME include a stateless multicast protocol, a Delaunay distributed hash table (DHT), as well as routing and host discovery protocols for a hierarchical network. ROME protocols do not use broadcast and provide both control-plane and data-plane scalability. Extensive experimental results from a packet-level event-driven simulator, in which ROME protocols are implemented in detail, show that ROME protocols are efficient and scalable to metropolitan size. Furthermore, ROME protocols are highly resilient to network dynamics. The routing latency of ROME is only slightly higher than shortest-path latency. To demonstrate scalability, we provide simulation performance results for ROME networks with up to 25 000 switches and 1.25 million hosts.