GMB operation
Earlier software releases supported GMB configuration on a per-port basis. Beginning with software release 15.18, the 2920 and 5400R switches also support GMB configuration on static trunks. (GMB configuration is not supported on dynamic LACP or distributed (DT) trunks.
For application to static trunk interfaces (2920 and 5400r only), GMB enforcement is applied individually to each port belonging to the trunk, and not to the trunk as a whole.
For any port, group of ports or, static trunks, you can use the default minimum bandwidth settings for each outbound priority queue or a customized bandwidth profile. It is also possible to disable the feature entirely.
The switch services per-port outbound traffic in a descending order of priority; that is, from the highest priority to the lowest priority. By default, each port (including each port in a static trunk) offers eight prioritized, outbound traffic queues. Tagged VLAN traffic is prioritized according to the 802.1p priority the traffic carries. Untagged VLAN traffic is assigned a priority of 0 (normal).
802.1p Priority settings in tagged VLAN packets1 |
Outbound priority queue for a given port |
|---|---|
1 (low) |
1 |
2 (low) |
2 |
0 (normal) |
3 |
3 (normal) |
4 |
4 (medium) |
5 |
5 (medium) |
6 |
6 (high) |
7 |
7 (high) |
8 |
The switch processes outbound traffic from an untagged port at the "0" (normal) priority level.
You can use GMB to reserve a specific percentage of each port's available outbound bandwidth for each of the eight priority queues. This means that regardless of the amount of high-priority outbound traffic on a port (including each port in a static trunk), you can ensure that there will always be bandwidth reserved for lower-priority traffic.
Since the switch services outbound traffic according to priority (highest to lowest), the highest-priority outbound traffic on a given port automatically receives the first priority in servicing. Thus, in most applications, it is necessary only to specify the minimum bandwidth you want to allocate to the lower priority queues. In this case, the high-priority traffic automatically receives all unassigned bandwidth without starving the lower-priority queues.
Conversely, configuring a bandwidth minimum on only the high-priority outbound queue of a port or static trunk (and not providing a bandwidth minimum for the lower-priority queues) is not recommended, because it may "starve" the lower-priority queues.
For a given port, when the demand on one or more outbound queues exceeds the minimum bandwidth configured for those queues, the switch apportions unallocated bandwidth to these queues on a priority basis. As a result, specifying a minimum bandwidth for a high-priority queue but not specifying a minimum for lower-priority queues can starve the lower-priority queues during periods of high demand on the high priority queue. For example, if a port or static trunk configured to allocate a minimum bandwidth of 80% for outbound high-priority traffic experiences a demand above this minimum, this burst starves lower-priority queues that do not have a minimum configured. Normally, this will not altogether halt lower priority traffic on the network, but will likely cause delays in the delivery of the lower-priority traffic.
The sum of the GMB settings for all outbound queues on a given port or static trunk cannot exceed 100%.