QoS-Policy on ASR1k Main Interface but not on Sub-Interface

[pille1234]

G'day everyone,

I have an ASR 1K where I have a hierarchical Qos policy doing sub linerate shaping on the main interface based on DSCP. That is working with no problem.

Now what happens if I add a sub-interface without any policy-map? Is the main shaping policy being applied to the sub-interface as well? Is the subinterface not using anything but the default class? Is the subinterface potentially harming the policy on the main interface?

I couldn't find any good documentation about that situation.

Regards

[pigallo]

Hello @pille1234 ,

please see my answers inline.

---

@pille1234 wrote:

G'day everyone,

I have an ASR 1K where I have a hierarchical Qos policy doing sub linerate shaping on the main interface based on DSCP. That is working with no problem.

Now what happens if I add a sub-interface without any policy-map? Is the main shaping policy being applied to the sub-interface as well?

ASR1k handles hierarchical QoS so, in your case, being the subif a subset of the main interface, it will rely on parent scheduling on egress side. If you have more subinterfaces relying on the same parent node, then there will be bandwidth contention among them if parent node is congested and it will be handled in function of bandwidth ratio of each sub-node.

Is the subinterface not using anything but the default class?

Question is not clear can you elaborate ?

Is the subinterface potentially harming the policy on the main interface?

No, the subif cannot arm parent policy as it doesn't uses any queuing/non queueing mechanisms.
 Feel free to ask more precise question about use case, as the term "harming" may be very vague in such case.

I couldn't find any good documentation about that situation.

Regards

---

[pille1234]

Hallo @pigallo,

thanks for your response.  I agree, there is not much documentation available for this situation.

I am not sure if I simply don't fully grasp your response or if there is a misunderstanding, so please let me clarify what I am asking. 

policy-map PARENT
class class default
shape average 60
service-policy CHILD

policy-map CHILD
 class EF
  priority percent 10
 class AF
  bandwidth percent 50
 class class-default
  random-detect dscp-based
  bandwidth percent 40

interface TenGigabitEthernet0/1/0
 service-policy output PARENT

So far I believe I understand what's happening: I have a given bandwidth, e.g. 10G and this is being shaped down to 6G with a prio queue and different minimum bandwidth reservations.

Now I add the subinterface  TenGigabitEthernet0/1/0.100 without adding any additional service policy. Is the traffic that passes this subinterface shaped to 5G and is the subsequent child policy applied to the subinterface traffic as well?

If not, is any traffic passing the subinterface accounted for in the main service policy or could we theoretically see 6G traffic on main interface plus 4G on the sub interface? Is the prio queue for the main interface taking precedence over the subinterface in case of contention?

Regards

[pigallo · Accepted solution]

Hi @pille1234 ,

---

Is the traffic that passes this subinterface shaped to 5G and is the subsequent child policy applied to the subinterface traffic as well?

The traffic under the subinterface in this case relies on what is configured, in terms of egress queueing and scheduling, on the main interface. So, if you configured 6Gbps shaper on the parent node, you cannot have more than 6Gbps traffic shaped on the single subnode, assuming that other subnodes, if any configured, are not transmitting any traffic at the moment. If you have more than one subnode transmitting at same time, let's say you have 3 subinterfaces - TenGigabitEthernet0/1/0.100 - TenGigabitEthernet0/1/0.101 - TenGigabitEthernet0/1/0.102, then all these three subinterfaces when transmitting together won't be allowed to overtake 6Gbps because of parent shaper on main interface.

If not, is any traffic passing the subinterface accounted for in the main service policy or could we theoretically see 6G traffic on main interface plus 4G on the sub interface?

Yes it is accounted. You should consider the parent node as the closest point to the hardware ring in terms of scheduling. So, if parent node sets a limit, total amount of traffic coming from subnodes cannot overtake that limit. This rule is also valid with other types of subnodes like tunnels for example.

Is the prio queue for the main interface taking precedence over the subinterface in case of contention?

see answers above

Regards

---

[pille1234]

Hi @pigallo ,

got it now, thank you.