Protocol Description

Standard link aggregation enables you to directly connect two systems through multiple ports. These ports are viewed as one logical link, and may be simultaneously in active mode. This situation provides higher capacity between the two systems and supports automatic fail over if any of the physical ports becomes unavailable. Typically, you can use link aggregation between tiers for higher capacity. To provide redundancy, multiple up-links are deployed between tiers. However, because of the spanning tree algorithm, only one of these links is active.

LLA on the other hand, enables link aggregation links from servers to terminate at two separate switches. This situation provides an increase in capacity and redundancy, supporting an active-active deployment model. Fail-over can occur rapidly, as there are no L2 or L3 protocol delays. On the server, the links are configured as Linux bonds or Oracle Solaris link aggregations.

In LLA, two switches are connected together directly through an ISL, a special trunk group between them. The LLA ports are connected to the server and are members of the LLA port-channels. The ISL port-channel is required to be in forwarding state on both the switches at all times. This is done by setting the spanning tree cost on the ISL port-channel lower than any other port in the system, and by choosing the uplink costs such that the sum of the ISL cost and the preferred uplink cost is less than that of other uplinks. Both switches maintain their identity as separate L2/L3 switching entities. The rest of the network sees them as two separate L2 switches with different addresses.

When designing a deployment with LLA, carefully examine the expected traffic patterns. The ISL port-channel should be provisioned with an adequate bandwidth. Traffic from the uplinks may always be distributed between the local link and the path crossing the ISL. Likewise, depending on the server side configuration, traffic may always be using both paths in the reverse direction.

With LLA enabled, the switches must share state and maintain a common mapping of ports. This common mapping is initialized when SEFOS starts up. For this reason, after LLA is configured, the configuration must be saved and SEFOS must be restarted.