5.5 Redundant Links
5.5.1 EtherChannel technology
Another mechanism for redundant links in a Spanning-Tree environment is called Fast EtherChannel technology. Fast EtherChannel technology builds upon standards-based 802.3 full-duplex Fast Ethernet. This technology allows parallel links to be treated by Spanning Tree as one physical link.

Fast EtherChannel technology offers bandwidth scalability within the campus by providing full-duplex bandwidth of 200 to 800 Mbps. The implementation of Fast EtherChannel technology, in addition to providing high bandwidth, also provides load sharing and redundancy. This technology provides load balancing and management of each link by distributing Unicast, multicast, and broadcast traffic across the links in the channel. In addition, Fast EtherChannel technology provides redundancy in the event of link failure. If a link is lost in a Fast EtherChannel, traffic is rerouted to one of the other links in less than milliseconds, and the convergence is transparent to the user.

Fast EtherChannel and Gigabit EtherChannel use a load distribution algorithm based on the destination MAC address.

A bundle is a group of links managed by the Fast EtherChannel process. Fast EtherChannel technology provides statistical load distribution of connections over multiple links in a bundle. If one link in the bundle fails, the Ethernet Bundle Controller (EBC) informs the Enhanced Address Recognition Logic (EARL) of the failure, and the EARL ages out all addresses learned on that link. This condition allows the EBC and the EARL to recalculate in hardware the source-destination address pair on a different link. When a source address queries and the destination responds, the address is relearned on a different link in the bundle.

Fail-over time is the time it takes for the new address to be relearned. Assuming that one packet sent by the source results in an instant response, fail-over takes place as quickly as 10 microseconds. Fail-over may take longer because of the windowing that the particular application uses to send packets before it expects an acknowledgment. Even then, relearning should not take more than a few milliseconds, and as a result, no application or session timeout will be seen due to link failure.

When one link fails, all traffic is relearned from that link and carried on other links in the Fast EtherChannel. However, when the link comes back, traffic will not necessarily be immediately relearned on the reinstated link. For example, suppose link 1 fails and is relearned on link 2. Now link 1 is reconnected with all other active links. Traffic previously on that link never reverts back. Traffic stays on link 2 (or the link the traffic was relearned on after link 1 failure) until it is aged naturally out of the content-addressable memory (CAM) table, in which case, when that address is relearned, it is learned again on link 1.