QOS and AutoQos

[Mokhalil82]

Hi

I have not deployed QOS before so had a few questions regarding it. I am deploying a new site consisting of a tiered model using a core and access switches. I want to deploy QOS. I am using 2960x access switches with 3850 core switches. We have multiple sites connecting into the MPLS. 

1) The other sites or the DC do not currently have any QOS configured. Could I still get benefits out of deploying QOS just for this site? 

2) Also as I do not have hardly much QOS experience, would auto qos be a good option?

3) and finally does qos only take effect if the links reach full utilisation?

Just to add my main focus for using QOS would be to ensure voice traffic is given priority

Thanks

[Mark Malone · Accepted solution]

Hey

1 Recommended QOS is deployed from source to destination right through your path start - end , not always possible but you should try if your going to deploy it

2 autoqos is fine on layer 2 switches , you don't really need to be building service policies for every switch an pure l2 switches wont allow them, make sure though your layer 3 devices where you break out has service policys applied too you don't want just your LAN running it, autoqos is based on hardware per platform while your layer 3 qos service policy is done in software

3 In reality yes its only useful when link is being hammered as theres not much point in prioritizing traffic if the theres no load

As well 3850s trust dscp by default and if your using Cisco phones it will automatically carry the trusted values through the switch without you having to do any configuration , it does have the option for software qos as well and it also can create full automated policy's

If its just for voice all you need is to set the access ports as mls qos trust dscp and the same on your uplinks , reason being this will cover you for L2 and L3 once you have mls qos map cos-dscp 0 8 16 24 32 46 48 56 due to the fact even if you have old phones that use cos instead of dscp the switch will map the values between them

Problem I ran into rolling this out globally was 2960 bucket and buffers struggle with certain traffic when full mls is globally enabled , we had to remove it from some 2960 where there were filers as bucket/buffers were being overloaded in certain hardware queues , now saying that you can manipulate the queues if required but I wouldn't personally  alter them too much , we shouldn't have filers on access switches either but not all remote sites get kitted out with the devices sometimes they really need:)

To test you can use a wireshark capture the port traffic and make sure the phone is sending out an EF 46 value for voice so its highest priority this confirmed access port then capture the uplinks make sure you see the same to confirm markings are being kept as passing through switch

Another test is to use an extended ping and set the tos value to voice think its 184 this will should show the counters increase in right queues

[Mokhalil82]

Thanks Mark for answering my questions. I will put that into practice and test on the network I am currently deploying

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Unfortunately, questions one and two are "it depends" kind of answers.

#1 Depends much on your traffic flows, and where there might be congestion.  Often, off-site links are bottlenecks, so if you're not going to deploy QoS everywhere, such interfaces often should be the first to have QoS configured on them.

You might get the most QoS benefit, just applying QoS to your WAN's (for egress) interface.

#2 Depends much whether your traffic and your traffic markings, match the auto QoS model.  The better the match, the better it works for you.  However, if your traffic doesn't correspond to what the auto QoS model was designed for, it can actually be adverse to your traffic.

In common usage, auto QoS tends to provide best service for DSCP EF (or CoS 5) traffic, and as often one or both markings are used by VoIP bearer traffic, auto QoS can be, by default, beneficial for that.

#3 yes and no, answer depends what you mean by full utilization.  QoS is used to manage congestion.  I define interface/link congestion, whenever a frame/packet cannot be transmitted immediately.  Congestion can be transient, so a link, and your traffic, might benefit from QoS even for "low" utilization.  Conversely, a link might be running at 100% utilization, but not need or benefit from QoS.  It all boils down to, when there's congestion (again, when a packet cannot be immediately transmitted), is the delay (or loss of frame/packet) adverse enough to the service needs of the traffic?  If it's not, you don't need QoS.  If it is, you either need more bandwidth, or you might need to manage your congestion, with QoS, such that different traffic types, meet their service needs.

For example, you mention VoIP.  Well, if you have large data transfer using FTP, and a VoIP call on the same link, and if the FTP transfer keeps trying to push past available bandwidth, the FTP traffic will likely cause service issues for the VoIP traffic.  If QoS is used to prioritize the VoIP traffic over the FTP traffic, then both traffic types are "happy".

An example of where utilization can be misleading.  If the VoIP call uses only 100 Kbps of a T1, for five minutes (assuming a typical SNMP 5 minute poll), your utilization is only about 7%.  If FTP comes up for just 10 seconds, your 5 minute utilization will barely change, but during that 10 seconds, your VoIP call's quality might totally tank.

Or, if you had about 15 concurrent VoIP calls, each 100 Kbps, on the same T1, your utilization would be close to 100%, but you shouldn't have any VoIP quality issues.