Floodless in SEATTLE: A Scalable Ethernet Architecture for Large Enterprises

Summary

This paper puts forth a new network architecture, SEATTLE, that combines the simplicity of Ethernet and the scalability of IP. SEATTLE achieves this with 3 main features. First, it uses a one-hop, network layer DHT, which doesn't require hosts to flood the network to disseminate host locations and instead has a global switch-level view that stores the location of each host. Second, it uses a traffic-driven location resolution and caching, which helps packets travel the shortest path to their destination. Finally, it uses a scalable, prompt cache-update protocol which uses unicast to inform other switches when a hash table entry is invalid, leading to faster reaction times.

The authors run several simulations using real traces supplemented with synthetic data. They use 4 topologies of various sizes. They find that SEATTLE uses much smaller tables, has 1000 times less control overhead, a smaller stretch than ROFL and reacts faster to network changes and host mobility. SEATTLE is also tested in a real deployment and the results were found to be comparable to those in the simulation.

Criticism & Questions

I think this was paper was well-written and conveyed the features included in SEATTLE well. I think this network architecture has real potential and would be interested in seeing how it performs when implemented in a more wide-scale deployment. As the authors mention, it would also be interesting to see how this architecture could be improved. One thing I would have liked to see is more information on the issues that would come up when implementing SEATTLE in practice.