Solved: priority queue at WAN edges?

prakadeesh · ‎02-02-2010

Hi Folks,

My test scenario is that voip traffic is going to traverse between one site to another. Mostly the voice gateways are split across the sites. Now in the setup 3750 are interconnecting the intersite links on a LES widely on 100Mbps. I was more worried about the priority queue in the c3750, its a strict priority queue, so the SRR will not service other queues until priority queue is free. Incase of an edge where its connecting to another site, inter site voice traffic can accumulate, so wouldnt the c3750 be constantly be servicing the priority queue, because lots of people might be on calls? Or have I understood the SRR strict priority wrong? Incase we disbale the priority queue , would just shape( weight ) command help in serviceing the voip traffic more quickly than data?

Thanks,

Prakadeesh

Jon Marshall · ‎02-02-2010

prakadeesh wrote:

Thank you so much Jon for your inputs. Yes I am also waiting for the info but just got this weird thinking when testing the setup.

"You allocate the bandwidth needed to cater for the number of calls, hence you need to know the number of calls, the codec etc. to work out how much bandwidth is needed." your quote refers to the actual WAN bandwith I assume, or is it to the SRR queue bandwidth. If its SRR queue bandwidth, say I need 10M for voice so I should use the srr-queue shape 10 0 0 0 to allocate 10M to the voip traffic. But priority queue out will ignore this command anyways, this is where exactly I am confused. Sorry

Thanks,

Prakadeesh

No need to apologise, catalyst QOS is not one of the easier subjects to understand

If you need 10M for VOIP then you need to rate-limit the amount of VOIP traffic on the ingress ie. coming into your switch so it cannot exceed 10M. You cannot do the rate-limiting on the egress interface because rate-limiting is not supported on egress. So you don't specify a bandwidth on the egress priority queue you limit the amount of priority traffic as it enters the switch.

Jon

View solution in original post

Jon Marshall · ‎02-02-2010

prakadeesh wrote:

Hi Folks,

My test scenario is that voip traffic is going to traverse between one site to another. Mostly the voice gateways are split across the sites. Now in the setup 3750 are interconnecting the intersite links on a LES widely on 100Mbps. I was more worried about the priority queue in the c3750, its a strict priority queue, so the SRR will not service other queues until priority queue is free. Incase of an edge where its connecting to another site, inter site voice traffic can accumulate, so wouldnt the c3750 be constantly be servicing the priority queue, because lots of people might be on calls? Or have I understood the SRR strict priority wrong? Incase we disbale the priority queue , would just shape( weight ) command help in serviceing the voip traffic more quickly than data?

Thanks,

Prakadeesh

You have understood the priority queue funtion correctly ie. it will be serviced until it is empty and then the other queues will be serviced. However that is exactly what you want with VOIP and indeed that is one of the main uses for the priority queue ie. voice traffic.

You should have an understanding of how many calls on average will be made within your network. Without knowing the traffic patterns you cannot correctly allocate the correct bandwidth, not just to VOIP but also to other traffic. If you truly do not know how much bandwidth your voice data is using then yes you have the potential to starve the other queues but being honest, you should not be allocating bandwidth if you do not know how much is needed.

Jon

prakadeesh · ‎02-02-2010

Hello Jon,

Thanks for the reply. Yes this setup is actually a pilot so I can gather info as much as possible. But if strict priority is enabled both shape and share commands or omitted in SRR right? So, I dont understand how a bandwidth can control the queue starvation? Please advice.

Thanks,

Prakadeesh

Jon Marshall · ‎02-02-2010

prakadeesh wrote:

Hello Jon,

Thanks for the reply. Yes this setup is actually a pilot so I can gather info as much as possible. But if strict priority is enabled both shape and share commands or omitted in SRR right? So, I dont understand how a bandwidth can control the queue starvation? Please advice.

Thanks,

Prakadeesh

If the strict priority queue is enabled then SRR and the bandwidth weights still applies to the other queues just not to the priority queue. My point is that if you did happen to generate 100Mpbs of voice traffic then you would indeed starve the other queues because the switch will always be servicing the priority queue. But you don't do this when rolling out QOS. You allocate the bandwidth needed to cater for the number of calls, hence you need to know the number of calls, the codec etc. to work out how much bandwidth is needed.

Just allocating all voice traffic to the priority queue without knowing how much traffic that is could be a recipe for disaster.

Jon

prakadeesh · ‎02-02-2010

each Voip call generates around 50 pps( 64 kbps) for G.711 codec. C3750 running cef is capable of 6 million pps. so ideally it will take given the bandwidth is 100mbps, 1562( 100M/64K) simultaneous phone calls to cause queue starvation, because this will cause a constant streaming of voip packets . hope I am correct.

Thanks,

Prakadeesh

Jon Marshall · ‎02-02-2010

prakadeesh wrote:

each Voip call generates around 50 pps( 64 kbps) for G.711 codec. C3750 running cef is capable of 6 million pps. so ideally it will take given the bandwidth is 100mbps, 1562( 100M/64K) simultaneous phone calls to cause queue starvation, because this will cause a constant streaming of voip packets . hope I am correct.

Thanks,

Prakadeesh

It's just occured to me that you might be asking how you can control the priority queue on egress when you cannot allocate how much bandwidth the priority queue is allowed to use.

If this is your question, firstly apologies for not understanding before and secondly you need to police traffic on ingress to the swtch so that the amount of traffic going into the egress priority queue does not starve all the other egress queues. You want VOIP traffic to get priority through the whole switch and because the 3750 does not support egress policing you need to do the limiting on the ingress.

Jon

prakadeesh · ‎02-02-2010

Thank you so much Jon for your inputs. Yes I am also waiting for the info but just got this weird thinking when testing the setup.

"You allocate the bandwidth needed to cater for the number of calls, hence you need to know the number of calls, the codec etc. to work out how much bandwidth is needed." your quote refers to the actual WAN bandwith I assume, or is it to the SRR queue bandwidth. If its SRR queue bandwidth, say I need 10M for voice so I should use the srr-queue shape 10 0 0 0 to allocate 10M to the voip traffic. But priority queue out will ignore this command anyways, this is where exactly I am confused. Sorry

Thanks,

Prakadeesh

Jon Marshall · ‎02-02-2010

prakadeesh wrote:

Thank you so much Jon for your inputs. Yes I am also waiting for the info but just got this weird thinking when testing the setup.

"You allocate the bandwidth needed to cater for the number of calls, hence you need to know the number of calls, the codec etc. to work out how much bandwidth is needed." your quote refers to the actual WAN bandwith I assume, or is it to the SRR queue bandwidth. If its SRR queue bandwidth, say I need 10M for voice so I should use the srr-queue shape 10 0 0 0 to allocate 10M to the voip traffic. But priority queue out will ignore this command anyways, this is where exactly I am confused. Sorry

Thanks,

Prakadeesh

No need to apologise, catalyst QOS is not one of the easier subjects to understand

If you need 10M for VOIP then you need to rate-limit the amount of VOIP traffic on the ingress ie. coming into your switch so it cannot exceed 10M. You cannot do the rate-limiting on the egress interface because rate-limiting is not supported on egress. So you don't specify a bandwidth on the egress priority queue you limit the amount of priority traffic as it enters the switch.

Jon

prakadeesh · ‎02-02-2010

Thanks Jon, you are a star . I will look into the ingress ratelimiting .