Flooding Traffic Reduction through Optimized Address Learning in Carrier-Grade Ethernet Networks

Inhalt:
As Ethernet deployments become larger and the number of active MAC addresses in the network grows, flooding (broadcast based resolution of the location of unknown addresses) becomes a scalability bottleneck. This is especially a problem during network convergence, e.g. after a spanning tree topology reconfiguration or a ring protection switch, which result in deletion of forwarding databases. Network performance may drop significantly during periods of flooding; a reason why these should be kept as short as possible. The limited learning bandwidth – due to its complexity the MAC address learning process is usually not designed for wire-speed performance – additionally prolongs a flooding phase. In this work, we qualitatively evaluate several means of improving the learning performance. We demonstrate that ranking the switch internal learning requests with respect to their ability to reduce flooding is a new way to mitigate the broadcast problem.