Understanding how unused bandwidth is distributed by bandwidth remaining ratio

[frangelico.eisma]

Hello all,

I have a question regarding how unused bandwidth is distributed in each class-map when bandwidth remaining ratio is configured. Let's take the following example below.

policy-map PARENT_POLICY
    class class-default
        shape average 100000000   
            service-policy CHILD_POLICY
   
   
            policy-map CHILD_POLICY
                class EF_CLASSMAP
                    police 50000000
                    priority
                class AF_CLASSMAP
                    police 30000000
                    bandwidth remaining ratio 9
                class class-default
                    bandwidth remaining ratio 1

Based on the simply configuration above, let us assume the following:

• If there are no other traffic except for class-default traffic.
• can the class-default traffic go up to 100000000 bitrate without dropping any packets?
• my understanding here is that the class-default can only go up until 10% of what is available and not being used by the other class.

I've look into this as my reference

http://www1.cisco.com/c/en/us/td/docs/ios-xml/ios/qos_plcshp/configuration/xe-16/qos-plcshp-xe-16-book/qos-plcshp-dist-rem-bw.pdf

Our problem with our network is that our provider is insisting that the class-default can take up as much unused bandwidth considering there are no other traffic for the other class. But in reality, we are experiencing that the class-default is already dropping packets even though there is no congestion yet on the link. So in context to the above example, our class-default is dropping packets even though we are at just about 50% of the total shape rate.

I'd appreciate your expert opinion and thanks in advance.

[Joseph W. Doherty]

The way I read the documentation of this feature, from your reference link, class-default should be allowed to obtain all unused bandwidth.

As to you seeing drops when you get to 50%, that may have more to due with your providers Tc and how deep your provider's shaper's queues are vs. the characteristics of your traffic.

[Adam Vitkovsky]

This is the fundamental difference between the Police, Shape and Bandwidth:

Police = allowed maximum for a given class all the times, whether there’s congestion or not

Shape = allowed maximum for a given class all the times, whether there’s congestion or not

Bandwith = guaranteed minimum for a given class in case there’s a congestion and different classes of traffic are fighting over available BW resources (active only during congestion)

So the configuration is saying that in case EF and AF and BE are causing congestion together, (dictated by the parent shaper –so sum of traffic is more than 100M), then when worse comes to worst and all three classes transmit at 100M then EF is going to be capped at 50M and AF traffic is going to be guaranteed at least 9 pieces of the remaining 50M and BE will be guaranteed at least 1 piece of the remaining 50M.

If however there’s no EF traffic then all 100M “remainder” is available for AF and BE; and AF and BE each transmit at 100M, then AF traffic is going to be guaranteed at least 9 pieces of the remaining 100M and BE will be guaranteed at least 1 piece of the remaining 100M.

So as you can see the config is just making sure that AF and BE gets at least their guaranteed share of the leftover BW after EF (marked as “priority”) class was served.

By the same token if EF and AF is not transmitting then all the leftover BW is available to just BE class.

For more info:

http://www.cisco.com/c/en/us/td/docs/routers/xr12000/software/xr12k_r4-2/qos/command/reference/b_qos_cr42xr12k/b_qos_cr42xr12k_chapter_01.html#wp1075143779

and this one has a very good example:

http://www.cisco.com/c/en/us/td/docs/routers/asr9000/software/asr9k_r5-1/qos/configuration/guide/b_qos_cg51xasr/b_qos_cg51xasr_chapter_0111.html#ID266

Using “show qos interface” or “show policy-map interface” you should see that there are no drops on the BE class.

adam

netconsultings.com

::carrier-class solutions for the telecommunications industry::

[Yaroslav Voloshkevich]

Thank you for the explanation, Adam.
I have a question though: in the second case when there's no EF traffic at all - how "AF traffic is going to be guaranteed at least 9 pieces of the remaining 100M" if it actually has the "police 30000000" configured together with "bandwidth remaining" command?
From the "Police = allowed maximum for a given class all the times" part I understood the traffic for AF can never go over 30000000, no matter what, and 9/10 of 100M will never be reached.

Could you elaborate on this, please?

Yarek