Priority command under Policy-map question

[Andres Franco]

Hello all.

First, the scenario:

switchB ------swichA ===== Router

                     |

                     |

                 switchC

I´m implementing QoS with some Catalyst 3560 (48xGE) switches and a 7600 router. I did classify and manage congestion without problem between switches, and traffic is reaching the router with the correct values. But once on the router, i have not applied any action. Since the traffic i´m managing is voice, i was thinking to classify and then issue the priority command under a policy-map for this. Phones and PBX are not Cisco (Nortel-Avaya instead). The problem is that i don´t know how to calculate the correct bandwith value that should follow the priority command. SwitchA has a port-channel (2 GbE interfaces on both sides) trunk connecting to the router. How could i calculate this value? any suggestion? Thanks in advance.

AF.

[Marwan ALshawi · Accepted solution]

Hi AF,

first of all bandwidth per cal is something relate to the voice compression/codec such as G711 or G729 as each codec has differnt amount of bandwidth per call for example for G711 you will need around 80K per call this is with L2 overhead added to it

so if you have 10 calles then 10x80K

however to get your QoS for voice working correctly you need to make sure it is end to end ( if possible) for example if you have a policy map with priority to service voice first in the case of link congestion then where the link/traffic gose after this is it service provider cloud ? if yes do you have agreement with the SP about the voice market traffic to honor the marking and treat it as priority within their cloud too ? if not then the QoS will be up to the router after the router the traffic within the SP cloud will be treated in a best effort

also you need to make sure that you shape your physical interface to the actual interface bandwidth if this is going over WAN for example you have Gig interface connected to a WAN link that is service up 10M then you need to shape that interface to 10Mbps then apply another QoS policy ( sub policy ) under the shaping one with the QoS Queuing

see the below article for more details:

https://supportforums.cisco.com/docs/DOC-8373

Hope this help

if helpful rate

[Andres Franco]

Hey marwanshawi, thanks for your reply.

I guess you did mistake when typing 8k when referring to G711 plus header stuff. But I think I got it, you mean, 64k + headers overhead, right? So it´s around 80k per call, so I could stimate around 3x46x80k of bandwidth for the 3 switches, right? . I understand about maintaining QoS end to end. Now, about maintaining end to end QoS, specially over WAN, what happens if my voice traffic will reach a remote destination over WAN using a MPLS link, and  I can manage the receiving router on the other end, will I need to ask the service provider to manage my marked traffic, or i could handle it by my own?. Thanks a lot for your help. (i´ll rate in the next round of answer, so don´t worry).

AF.

[Marwan ALshawi · Accepted solution]

Hi AF

ya i meant 80K

and yes you need to agree with your service provide about the QoS and class of service within their cloud as it is something require additional payment normally and just make sure you send the traffic with the right marking ( agreed with the SP ) for end to end QoS

for example if your voice market as DSCP EF and the service provider support that marking and agreed to be placed in the priority queue then you expect to receive voice traffic on the remote site marked as EF ( end to end )

if its differnt marking in the cloud for example ( which is a bit harder config here ) you need to remark the voice traffic when being sent out to the marking used in the SP cloud and remarked back to EF in the remote site inbound direction

Hope this help

[Andres Franco]

I guess i could use a dscp mutation or a policy-map to remark traffic. Ok, i have one last question. I was looking at a config example somewhere, and one guy did put the following (to be a applied on a 3560 access port):

policy-map QOS-3560

class VOICE

set dscp ef

police 128000 8000 exceed-action drop

! Voice traffic should not exceed 1 phone call

! excess beyond 128k will be dropped

Why did this guy said tha a voice call will require 128k??? i understood that without compression (G711) would be 64kbps + 16 kbps of header = 80 kbps, so did he include any kind of additional overhead??. Thanks.

[Marwan ALshawi]

no need to police it at 128 however you could put it as 100K just to be 100% covering any additional overhead things with your call

however this is only on the switch port

as you are not necessarily need 100K as mentioned above with G711 around 80K per call

however differnt layer 2 transport has differnt over head see the below link for more details

http://www.cisco.com/en/US/tech/tk652/tk698/technologies_tech_note09186a0080094ae2.shtml

hope this help

[Aaron Harrison]

Hi

It's also worth considering other scenarios rather than a basic call - for example if something uses the built-in-bridge or if you are using SIP based recording then you might have two voice streams at once from a phone, and this would be > 100K total.

You don't need to be overly strict with the priority statement, it's just low enough that it can't be 'abused' easily.

Aaron