Hi we have a requirement to restrict the bandwidth on the 10G link as it will be split in to two sub interfaces with one set to use 40% and the other of 60 % of the bandwidth. But wanted some pointers of how to do it and how to calculate the burst rate and the committed rates. as I found this config below on line but could do with some points. of how to adjust it for our situation, or if you have a better configuration suggestion.
Router(config)# class-map acgroup2
Router(config-cmap)# match access-group 2
Router(config)# policy-map police
Router(config-pmap)# class acgroup2
Router(config-pmap-c)# police 8000 2000 4000 conform-action transmit exceed-action set-qos-transmit 4 violate-action drop
Router(config)# interface fastethernet1/1/1
Router(config-if)# service-policy input police
try to avoid policing as much as possible and use shaping. Below is a sample config you can use. I somewhere remember that on the ASR 1000 you used to need to configure:
platform qos marker-statistics
Not sure if that is still necessary...
shape average percent 40 be 300 ms bc 400 ms
shape average percent 60 be 300 ms bc 400 ms
service-policy output SUBINT_40
service-policy output SUBINT_60
Also would the config need to be placed on the ingress or the egress on the router, as
The setup is as below, so would I need to add it to the 1Gb links, the ingress of the ASR router or the 10Gb sub interface egress. As right now the 10Gb line is actually a 1gb point to point link but we are planning to change it to a 10Gb link in the future.
switch1# to 1Gb ASR1# 10GB to ASR2# 1Gb to switch 3#
switch2# to 1Gb ASR1# 10GB to ASR2# 1Gb to switch 3#
" But wanted some pointers of how to do it and how to calculate the burst rate and the committed rates."
Imagine an interface of some fixed bandwidth that you want to allocate a certain amount of FIFO queue space; that's what you want to allocate for you burst size. (If you're wondering how much space you should allocate to an interface's FIFO queue, well that's an "it depends" answer.)
If you're asking about using a peak rate policer rather than an average rate policer (i.e. Bc+Be vs. just Bc), well often Cisco's peak rate policers often operate much like an average rate policer where the Bc of the latter equals the sum of the former's Bc and Be. I.e. it doesn't really make a difference. However, if you find a peak rate policer that actually "borrows" from Be than you can size Bc for the average desired and size Be to support microbursts. (If the foregoing appears to be somewhat complicated, yea it can be. So much so, using a shaper, as Georg suggests, is probably a better choice.)