Re: [rtcweb] DSCP and media

As Bernard points out there is the qos marking for the wireless networks and local networks but what I am mostly interested in is the device (typically an NAT) that is sequencing the packets on to the broadband uplink from my home to my ISP. Lots of NATs support some simple set of priorities and send the higher priority packets ahead of the lower priority. My view is this was largely done for the online gaming people but in many cases it does help voip. Many of the voip devices do take advantage of it - there is pretty much no downside for setting the DSCP. The cases where I see it helping are things like the following: I have a 128Kbps uplink, I am trying to send an 80Kbps audio stream, and at the same time I email a 25 MB powerpoint file. With QoS, typically the audio gets it's 80k and email gets the rest. WIthout QOS, the audio gets packet loss.  It helps in cases where the uplink is the congested part of the network. It does not help if your downlink is the problem. Keep in mind most people have much faster downlink than uplink. It's easy to run your own experiments and try this out - just get a nat with support and turn it on and off and see how things go. It also helps with wireless but I have not done a lot of experimenting with that - my wireless is so much faster than my broadband that it has not been a problem but I heard rumors that it can help file transfer over the lan not impact voip - I just don't know as much about this sort of use case. 

To your point about too risky to expose a generic API. I'm on the fence. You could be right but it's sort of hard to decide why if it too risking to let some JS send an packet with higher priority, that it is OK to let is send normal priority ones. It would be interesting to talk about the risk, and what we might do to mitigate them. One point of view is that rate shaping and policing of packets can never be done by the endpoint and needs to be done by network router and switches. The people that buy into this don't see it as any worse that the JS can set the DSCP than the fact that a think app on my notebook computer can set the DSCP.  

The other issue is that it is not clear what values you should set the DSCP to. If you read the appropriate RFCs, you will come to the conclusion that they are pretty confused on the topic. If you look at the values being used by Cisco, the mapping from 802.11, and Microsoft, you would notice they are not the same and there may be some uh "gaming the system" going on. One of the things this WG could do would be to come up with some clear guidance on what values to use for browsers and when to use them. 

On Jul 20, 2011, at 12:39 , Justin Uberti wrote:

> Can you go into more detail about what these NATs do with the DSCP markings?
> 
> I'm not convinced we want to expose a generic API for this (too easy to misuse), but I think it definitely makes sense for implementations to set appropriate markings on audio and video packets, including different levels on individual video packets when using hierarchical prediction structures.
> 
> On Wed, Jul 20, 2011 at 1:59 PM, Cullen Jennings <fluffy@cisco.com> wrote:
> 
> Many home NATs support DSCP based QoS and it does help. I think that the IETF should recommend that some default DSCP for audio and video from the browser as well as suggest there should be an API to set the DSCP for different media steam.
> 
> Cullen <with my individual contributor hat on>
> 
> _______________________________________________
> rtcweb mailing list
> rtcweb@ietf.org
> https://www.ietf.org/mailman/listinfo/rtcweb
> 

Cullen Jennings
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html