Data Center Fault-Tolerant Routing and Forwarding: An Approach based on Encoded Paths

Traditional layer 2 and layer 3 network designs face some limitations on data center networks such as lack of scalability, difficult management and inflexible communication. We observe that data center networks are often managed as a single logical network fabric with a well-controlled baseline topology and growth model. This paper describes a data center networking approach based on encoded paths carried in the packet header. We present an OpenFlow-based testbed implementation of a data center architecture governed by a logically centralized Network Manager, which is responsible to transparently provide the networking and support functions to operate the data center network. We evaluate the proposal in terms of fail recovery time, state requirements, and load balancing capabilities. The results of the experiments show that our proposal improves the fail recovery time, while preserving scalability requirements.