A centralized controller to improve fault tolerance in TRILL-based fabric networks

Fault tolerance is a challenging requirement in virtualized data centers. While distributed layer 2 routing TRILL-like protocols are robust and scalable, they still lack efficiency upon a link failure. In this paper, we propose a hybrid architecture that achieves the best tradeoff between efficiency of centralized architectures and robustness of distributed ones. This architecture is based on a centralized controller that ephemerally configures the routing table with alternative paths already computed by the TRILL IS-IS protocol. This ephemeral entry enables service continuity while the TRILL. IS-IS recomputes the definitive route. The experimental evaluation shows that the recovery from the link failure is almost instantanious.