QoS for Video over IP

[d.hodgson]

Hi folks,

we are looking at installing video over IP throughout our infrastructure and will be trialing it soon. We have been told by the supplier that it will need about 2mbps bandwidth, so we have arranged to have 2 of our existing links to be increased by 2Mbps. What we'd like to do is to prioritise and limit the video over IP traffic to 2Mbps to verify that it is indeed ok to run on 2Mbps. For the trial only we don't want the Video over IP traffic to go over 2 Mbps.

I've attached our QoS settings on our core router and was wondering if you kind folks could fill in the gaps for video over IP? I'm trying to find out DSCP values and stuff so place an XX or something instead of what will be the real value if that makes sense?

I've looked at "shape peak" or "police CIR" but I don't understand both enough to decide if any will suite what we want.

thanks for your time and patience.

Dave

[mannagar]

Hello Dave,

Please refer this link..

http://www.cisco.com/en/US/tech/tk543/tk757/technologies_tech_note09186a0080094968.shtml.

hope this help you,

Mano

[Joseph W. Doherty]

You didn't describe the type of video. If it's some kind of streaming video, setting both your video class's bandwidth to 2 Mbps and an average shaper to 2 Mbps might be close to want you want. (Note: on most platforms, I believe FQ within class-default can preclude your other non-LLQ class bandwidth guarantees.)

e.g. (NB: syntax might be incorrect)

class four

set ip precedence 4

bandwidth percent 25 (8 Mbps PVC?)

shape average percent 25 (percent supported later IOSs)

If the video is real-time, like video conferencing, I believe it belongs in LLQ. You can define another priority LLQ class with its bandwidth set to, perhaps, 3 Mbps, and with a peak shaper set to, perhaps, 2 Mbps.

e.g.

class four

set ip precedence 4

priority percent 38 (8 Mbps PVC?)

shape peak percent 25 (percent supported later IOSs)

Markings don't really matter unless you're going to use them later downstream. If it's real-time video, normally you would mark with IP Precedence 4 or AF4x marking.

[d.hodgson]

Hi Joseph,

Sorry, it is Video Conferencing and it's between our sites. It's multipoint with the hub being at our Perth office. The test will be a video conference between one remote site and our Perth office.

In your second example would that mean that it can LLQ 3 Mbps but then every thing after that is normal priority and the total amount of bandwidth on the WAN that can be used at any time (non-congested and congested) will be 2Mbps (shape peak percent 25)? If that's so why the 3Mbps?

thanks

Dave

[Joseph W. Doherty]

I chose 3 Mbps for the LLQ limit in the second example both because the shaper should control the rate at 25% and to help insure LLQ policer doesn't drop packets (i.e. they're shaped). If you set both LLQ and the shaper to the same bandwidth, then the question arives, which deals with over rate burst packets first, the shaper or the LLQ policer? Also, when you define an explicit shaper (or policer) you can often set Bc and Be, but don't believe that's possible with the implicit LLQ policer. If the LLQ's Bc and Be are smaller than the explict shaper's, I would expect it to drop packets.

In the first example, where an explicit shaper is being used with an ordinary bandwidth class statement, the latter doesn't police packets when they go over rate, so there's no conflict.

[d.hodgson]

wow, now I see why I need to complete my CCNP quickly. Thanks for the info it now makes sense why you chose 3Mbps. But are you telling me that the Video conference can burst to 3Mbps using your config? I'd really like it to reach a maximum of 2Mbps only including bursts.

thanks

Dave

[Joseph W. Doherty]

Actual bursting of your video will depend much on its codec, but many are very bursty.

The shaper will smooth bursts. To keep the average video bandwidth as close as possible to an actual 2 Mbps, use a minimum Tc (4 ms is min, I believe, on many platforms).

[edit]

Also use an average shaper.

[d.hodgson]

Sorry to keep bringing up something you may have answered already but are you saying I cannot limit the amount of bandwidth assigned to VC to 2Mbps? Are you saying I can only shape it so it smooths out peaks to "average" of 2Mbps?

thanks

Dave

[Joseph W. Doherty · Accepted solution]

Actual tranmission is always at the speed of the interface. A shaper may queue some packets to maintain an average rate, a policer may drop somes packets to maintain an average rate. So yes the shaper will limit the bandwidth to your desired bandwidth of 2 Mbps but based on an average. For what you're trying to accomplish, this should be "good enough".

[d.hodgson]

Superb, I'll set this up and see how we go. Many thanks for your help and patience Joseph.

Dave