Auto QoS & WRR

Jun 25, 2005 3 Replies

Got a couple of questions regarding WRR on a 3550.



The following is added by AutoQoS globally:



mls qos map cos-dscp 0 8 16 26 32 46 48 56 mls qos min-reserve 5 170 mls qos min-reserve 6 85 mls qos min-reserve 7 51 mls qos min-reserve 8 34



The following is entered under the interface:



wrr-queue bandwidth 10 20 70 1 wrr-queue min-reserve 1 5 wrr-queue min-reserve 2 6 wrr-queue min-reserve 3 7 wrr-queue min-reserve 4 8 wrr-queue cos-map 1 0 1 wrr-queue cos-map 2 2 4 wrr-queue cos-map 3 3 6 7 wrr-queue cos-map 4 5



My questions are:



With respect to the 'wrr-queue bandwidth 10 20 70 1,' are these the percentages for each of the four output queues? If so, why is Queue 4 only getting 1% of the bandwidth (especially considering it is the priority queue)?



For the 'mls qos min-reserve' statements, what do the 5, 6, 7 and 8 refer to?


Jonathan,

The queues on a 3550 are defined as 1 priority and 3 WRR queues. The priority queue is preemptive and does not participate in the WRR scheme. Therefore the number in this example is pretty much useless. Actually, a 0 would make more sense to me. As you will notice, the other 3 queues add up to 100 percent.

The min-reserve relates to minimum queue buffer size. Earlier in your config, you should have commands like: mls qos min-reserve 5 170 mls qos min-reserve 6 10 mls qos min-reserve 7 65 mls qos min-reserve 8 26

There are 8 min-reserve values - all default to 100 packets. You can modify you minimum buffer size by setting the these and then applying to the queue in the interface.

Hope that helps,

Jim

So, why is the priority queue (queue 4) set for '1' by auto qos? Should it be changed to 0?

Well, according to Cisco, these aren't percentages, but ratios of service (well, 'bandwdith weghts')...

Do these correspond to CoS? But that doesn't make sense, so VoIP traffic would be given a larger queue depth, too large in my opinion as it would add ridiculous amounts of delay (however, if it is coupled into the priority queue, this queue should never actually buffer.

And, if the default queue depth is 100, then CoS 0 traffic will have less chance of tail drop than CoS 6, 7 and 8.

My question relates to poor voice quality for quite a few customers...

Given the above config, what changes would improve things?

Jonathan

My question s

That would make more sense to me. But, who knows why they did it that way.

Sounds like semantics to me.

Sort of.... The wrr-queue cos-map commands under the interface determine which queue to place the data in based on CoS. There are a lot of components to Qos and queueing. You need to make sure that your queues are set right and then place your traffic in the right queue. There are definitely trade off when setting buffer size. Too large and you do risk latency issues, but too small and you'll start getting drops. Even though the priority queue goes first, you can still get buffered data. If you have a lot of priority packets (lots of calls), some data will have to be buffered.

I wish there was an easy answer to this, but there is not. I think Cisco has designed the auto qos around what would work well for many customers. Every situation is different. You'd need to spend time investigating what is happening in that switch as well as the rest of the network. You can probably start by searching (Google and Cisco) for the issues that you are having and see what is the general cause of them - and what the suggested solutions are. Perhaps buffer overruns could be part of it. Perhaps your traffic is not getting tagged right. There are a lot of reasons and a lot of possible symptoms you could be having.

Good luck researching :-)

Jim

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required