cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
671
Views
0
Helpful
6
Replies

When QoS is enabled, network application performance is horrible

support
Level 1
Level 1

I'm in the process of planning a VoIP deployment on my network. The other day, I enabled QoS of one of my 3750 access switches so I could start figuring out what changes needed to be made to support VoIP. Here are the details.

C3750 Software (C3750-I9-M), Version 12.2(20)SE4 is what IOS I'm running, and the only commands for QoS I entered into the switch were:

mls qos

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

Right now, no traffic is marked or classified, but when I enabled QoS on the switch with the "mls qos" command, I noticed that all my network based application take a VERY noticeable nose down in the gutter with regards to performance. I even notice that it takes over a second to save a simple word document to the network. Once QoS is disabled, it behaves as expected.

Based on what I've read, the switch's default settings for QoS shouldn't treat unmarked traffic any differently unless I mark or trust the incoming traffic's CoS or DCSP values.

Can anyone point me in the right direction to troubleshoot this issue? In the end, I want default traffic to not suffer with performance issues, but I want voice traffic to use the “priority out” option when a call is made from a VoIP phone.

Thanks!

1 Accepted Solution

Accepted Solutions

I think you are running into the default behaviour of a 3750 (plus 2960 & 3560) where the default buffer sizes are not adequate for your network traffic and traffic is being dropped. Use the command 'show platform port-asic stats drop FastEthernet x/x/x' to see if you are seeing egress drops in any of the queues.

This is the default behaviour but it can be tuned, however IOS earlier than 12.2(25)SEE1 has some restrictions on how much buffer space can be allocated. See bug ID CSCsc96037.

I use the following global QoS settings in my 3560/3750 switch template, however you need at least 12.2(25)SEE1:

mls qos queue-set output 1 threshold 1 800 800 50 3200

mls qos queue-set output 1 threshold 2 560 640 100 800

mls qos queue-set output 1 threshold 3 800 800 50 3200

mls qos queue-set output 1 threshold 4 320 800 100 800

There are obviously other settings for DSCP-to-Queue and CoS-to-Queue that should be changed, however these settings change the egress queue thresholds for queueset 1 (the queueset all ports are mapped to by default).

HTH

Andy

View solution in original post

6 Replies 6

brichbou1
Level 1
Level 1

Found bug CSCeg29704 and I think that is my issue.

I think you are running into the default behaviour of a 3750 (plus 2960 & 3560) where the default buffer sizes are not adequate for your network traffic and traffic is being dropped. Use the command 'show platform port-asic stats drop FastEthernet x/x/x' to see if you are seeing egress drops in any of the queues.

This is the default behaviour but it can be tuned, however IOS earlier than 12.2(25)SEE1 has some restrictions on how much buffer space can be allocated. See bug ID CSCsc96037.

I use the following global QoS settings in my 3560/3750 switch template, however you need at least 12.2(25)SEE1:

mls qos queue-set output 1 threshold 1 800 800 50 3200

mls qos queue-set output 1 threshold 2 560 640 100 800

mls qos queue-set output 1 threshold 3 800 800 50 3200

mls qos queue-set output 1 threshold 4 320 800 100 800

There are obviously other settings for DSCP-to-Queue and CoS-to-Queue that should be changed, however these settings change the egress queue thresholds for queueset 1 (the queueset all ports are mapped to by default).

HTH

Andy

Andy,

I did discover that there was a bug in the IOS I was running too. I've been able to modify the thresholds and buffer sizes to get the traffic in queue 2 to behave as expected. I am also planning on upgrading the version of IOS on all my switches prior to deploying VoIP.

THe bug I found was different than the one you posted, but I will have to check that out first thing Monday morning.

Thanks for your help!

Hello, Andy!

We have similar problems. And there must be more people facing this.

Where did came up with your numbers? I mean, to me it seems like a major Cisco bug since Cisco themself says to not tweak the numbers.

Therefor, how did you came to the conclusion with your numbers? And what is your queue-settings. Could you please post your whole QoS-config.

Best Regards

Robert Maras

Hello, Andy!

How did end up with the numbers you suggest in your posting? And could you please post your entire QoS-config?

Best Regards

Robert Maras

Hi Robert

I can't really post entire configurations as this starts to encroach on intellectual property rights.... As a company we have tested and created various scripts that we use as a baseline. We then modify these after discussions and testing with customers during the design stage of a project.

As for why we have these queue-set values in the script - this is based on the Cisco QoS SRND version 3.3 design guide (which goes hand-in-hand with the End-to-End QoS Network Design CiscoPress book) however the values detailed in both of these are multiplied by eight. This takes into account the new values allowed by IOS release 12.2(25)SEE1 and later.

HTH

Andy

Getting Started

Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community:

Innovations in Cisco Full Stack Observability - A new webinar from Cisco