ANNOUNCEMENT - The community will be down for maintenace this Thursday August 13 from 12:00 AM PT to 02:00 AM PT. As a precaution save your work.
Showing results for 
Search instead for 
Did you mean: 

Relationship between VOQ and Queues created with MQC / ASR9K


I have got a question about the queuing structure on.

On a Fabric Interface Asic (FIA) of an ASR9K resides 3 VOQs for each egress port on other line cards.

...but ONLY 3 VOQs...


What if, when I have consistently structured my traffic into 8 classes (CS7, CS6, EF, AF4x, AF3x, AF2x, AF1x, DF) and dedicated egress queues with congestion avoidance mechanisms (if reasonable)?

In this case traffic of different classes with the same egress port destination using the 3 VOQs competitively and the little count of VOQs would be the weakest link of a chain of an end to end QoS concept.

I think I misunderstand some details of the VOQ concept and I would be deeply grateful for any advice.


Cisco Employee

Re: Relationship between VOQ and Queues created with MQC / ASR9K

Let me ask you this, how is a system (Cisco, Juniper, etc) supposed to know the priority of a packet to send it through any fabric to another LC?

We cannot rely on policing or shaping or bandwidth commands, we must mark packets with the "priority" keyword for those which we want to have priority and get serviced first. This is true outside of the fabric too.


For VoQs we have P1, P2, and BE (best effort) queues. At the ingress use a policy map to select "priority level" of the packet for those you want to have a higher priority, except control plane packets as those will be in the high priority queue already.


Also remember that the amount of time a packet sits in the fabric is typically measured in micro-seconds, so the difference in the queues is very little.





Re: Relationship between VOQ and Queues created with MQC / ASR9K

the packets which enter the ASR9K comes from the customer edge device (ASR920, ASR903 or ME1200) and are classified and marked there.
So, these packets come with a broad range of CoS or DSCP marking, but there are "only" two priority queues and one BE queue.
Because of this, all packets with AF4x down to BE marking with the same egress destination use the only available BE-VOQ?
If this right, the differentiated class handling would get lost, if packets belonging to different classes are handled by one single queue?
Content for Community-Ad