QoS design difference between Auto qos and cisco SRND

okuehn · ‎09-07-2007

we are going to roll out a cisco VoIP solution. our access layer switches are catalyst 3750 with advancedipservices IOS image, we use cisco 7940 ip phones.

actually we are struggling whether to use auto-qos configuration or following the cisco design guide "Enterprise QoS Solution Reference Network Design Guide Version 3.3"

i thought the configuration would be pretty the same but we found the following differences:

interface FastEthernet1/0/5

description IP PHONE CONFIGURED WITH AUTO QOS

switchport access vlan 70

switchport voice vlan 71

srr-queue bandwidth share 10 10 60 20

srr-queue bandwidth shape 10 0 0 0

mls qos trust device cisco-phone

mls qos trust cos

auto qos voip cisco-phone

spanning-tree portfast

end

!

interface FastEthernet1/0/6

description IP PHONE CONFIGURED WITH SRND

switchport access vlan 70

switchport voice vlan 71

srr-queue bandwidth share 1 70 25 5

srr-queue bandwidth shape 3 0 0 0

priority-queue out

mls qos trust device cisco-phone

mls qos trust cos

spanning-tree portfast

!

why is the auto qos interface missing the "priority-queue out" command? i think this turns queue 1 into a LLQ queue?

what is the reason for the different bandwidth allocations of the 4 shared queues?

10 10 60 20 compared to 1 70 25 5

one last question: cisco recommends to classify and mark traffic as close as possible to the device. if we have to classify an application according to layer 3 ip/port information, do we have to place an inbound policy-map to the layer2 access switch ports of the PC's ?

or should we keep our current configuration of marking traffic at the WAN aggregation routers by using NBAR?

many thanks in advance!

szahid · ‎09-07-2007

Pls see inline .

why is the auto qos interface missing the "priority-queue out" command? i think this turns queue 1 into a LLQ queue?

[Salman]

This has been true for auto-qos on 3750 since day 1. Auto-qos on a 3750 ( and for that matter 3560 , 2970 and 2960 ) will never enable the priority-queue out. That command does convert the Egress queue # 1 into a PQ.

what is the reason for the different bandwidth allocations of the 4 shared queues?

10 10 60 20 compared to 1 70 25 5

[Salman]

Its a possibility that SRND is written for a different code. The whole idea is to give the maximum bandwidth share to a queue to which your data traffic is mapped ( dscp = 0 ). If your data is mapped to queue 2 , then go with the number 1 70 25 5 and if your data is mapped to queue 3 , then go with 10 10 60 20. You can check which queue dscp(s) map to by using the command " show mls qos maps "

one last question: cisco recommends to classify and mark traffic as close as possible to the device. if we have to classify an application according to layer 3 ip/port information, do we have to place an inbound policy-map to the layer2 access switch ports of the PC's ?

or should we keep our current configuration of marking traffic at the WAN aggregation routers by using NBAR

[salman]

Just because a port is L2 will not restrict you from classifying the traffic based on L3 information. All switches these days have the capability of classifying traffic based on several different types of criteria including L3 info , dscp , vlan etc. And if the switch allows you to do that , it will be a good idea to move the classification to the edge.

thanks

Salman.

okuehn · ‎09-10-2007

Hi Salman, thanks for your reply.

following cisco's approach for using PQ/LLQ on real time traffic like voice, would you suggest to convert egress queue 1 into a PQ and enable priority queuing on the switch interfaces?

after looking the cos/dscp-queue mapping of Auto-Qos and SRND i understand the differences in the share queues: according to SRND, Q2 representing the Critical Data queue, Q3 representing the Best Effort queue (dscp/cos=0), and Q4 representing the Scavenger/Bulk queue. whereas in auto-qos the critical data is Q3 while besteffort and bulk remain both in Q4. so the two approaches utilize 60/70 percentage of the bandwidth for critical data, while only 20 percent is assigned to besteffort class. this sounds reasonable and seems to be pretty compareable.

however, i would suggest cisco to implement their default/auto-qos configuration according to their own best practice design guides.

thanks, Oliver

Joseph W. Doherty · ‎09-10-2007

"however, i would suggest cisco to implement their default/auto-qos configuration according to their own best practice design guides. "

Consider:

Best practices often evolve, so this also raises the question should the Auto-Qos change with new code releases?

Also, Auto-QoS, I think, is provided as an easy method to implement some QoS. Whatever it does, it's unlikely one size will fit all.

felixdav · ‎09-15-2007

Ther is an AutoQoS VoIP enhancement in 12.2(40)SE.

http://www.cisco.com/en/US/products/hw/switches/ps5023/prod_release_note09186a008089a797.html#wp850774

okuehn · ‎09-19-2007

yes, i've just installed 12.2(40) and configured auto qos cisco-phone on fe port. cisco has included the "priority-queue out" command in their current auto qos feature!

now the auto qos configuration correspond to their SRND.

they further unified their cos-dscp mapping from

mls qos map cos-dscp 0 8 16 26 32 46 48 56

to

mls qos map cos-dscp 0 8 16 24 32 46 48 56

thanks!