MRP/MSRP LAG Support

LAG is supported for MRP/MSRP. A LAG may have one or multiple member ports, thus providing redundancy as well as additional bandwidth.

There are two modes of how a LAG runs with MRP/MSRP. The two MRP/MSRP LAG modes are single-port and cumulative. The concept of effective bandwidth is used below to simulate the available bandwidth of a physical port (port speed). As in the case of the physical port, a configurable percentage, deltaBandwidth, which by default is 75%, of the effective bandwidth is the final bandwidth available for MSRP streams.

In single-port mode, the LAG is simply a way to provide redundancy. Therefore the effective bandwidth is set to the minimum bandwidth of all member ports.

effective bw = min (bw of all active member ports)

MSRP reservable bw = deltalBandwidth * effective bw

All LAG member ports on ExtremeXOS must have the same speed to aggregate. Therefore, the minimum bandwidth is equivalent to the bandwidth of any member port.

In cumulative mode, the LAG trades redundancy for extra bandwidth. To calculate the available bandwidth, all bandwidth of one active member and partial bandwidth of other member ports are used to calculate the effective bandwidth.

effective bw = min (bw of all active member ports) + beta * sum (bw of all other active member ports)

MSRP reservable bw = deltalBandwidth * effective bw

For example, if the LAG has three 1GB member ports, and deltaBandwidth = 75%, then in the single-port mode, the effective bandwidth is 1GB, of which 75%, that is 0.75GB is reservable for MSRP, exactly the same as in the case of a physical port.

For the same LAG in the cumulative mode, if beta = 50%, then 100% of one member port and 50% bandwidth of other two member ports contributes to the effective bandwidth, therefore:

effective bw = 1GB (one active member port) + 50% * 1GB (the next active member port) + 50% * 1GB (the last active member port) = 2GB

MSRP reservable bw = 75% x effective bw = 1.5GB

When a LAG runs in the cumulative mode, the streams are roughly evenly distributed on all member links. Even though beta percent bandwidth of the other member ports contributes to the effective bandwidth, it does not mean that reservation on that member port is limited to beta percent of its bandwidth. That is just the way to calculate an estimate of how much bandwidth can be provided with a LAG, with a balanced redundancy requirement. All member ports are treated equally and programmed in the same way.

As in the case of physical ports, if the (total requested BW) <= deltaBandwidth * (effective BW), then assume the LAG is able to handle all streams safely and provide a comfortable level of redundancy if the streams are reasonably evenly distributed. In extremely polarized cases, where many streams are hashed to a certain member link, packet drop is inevitable. When (total requested BW) > deltaBandwidth * (effective BW), then the LAG is not able to safely handle all streams, even though it may still be able to. At this time new reservation requests are declined.

The effective bandwidth is an aggregate of multiple physical ports and can exceed the bandwidth of any single physical port. The QoS profile is configured on a physical port, so if the reservation bandwidth request is less than the delta bandwidth of the physical port, the QoS profile will be configured to the requested bandwidth. To prevent the QoS profile configuration from exceeding the physical port bandwidth, the QoS profile configuration is limited to delta bandwidth of the physical port (which is typically 75% of the port bandwidth).