Does anyone know how CBWFQ scheduling really works - Cisco doesn't seem to share the low level details on this.
I have a situation where the class class-default is being starved by other queues.
I'm flooding a link with simulated UDP traffic that I have defined separate queues for (i.e. not in the class-default queue) and have reserved 25% of the link bandwidth for. This non class-default traffic is consuming almost 100% of the link. The traffic in the class-default queue is effectively being starved.
I have Flow based WFQ and WRED enabled on the class-default queue. But have tried all combinations of this without success.
According to Cisco docs the max-reserved-bandwidth 75, should ensure that 25% of the bandwidth is reserved for the class-default queue. This doesn't appear to work from my testing.
It is possible to disable WFQ in the class-default queue and define "bandwidth [kbps]" for the queue. However WFQ is a far better choice for the default queue as it will still consider DSCP.
Has anyone experienced this? I'm testing on 12.2(29).
