Radhika,
Thanks for the input which I welcome as I will unfortunately not be present at Portland.
Let me ask a few questions and make a few comments hopefully with the intent of opening up the debate.
1. I am not sure I understand your concept of a mapping table between the H.323 QOS and the transport layer QoS. My understanding is that QoS is on three levels:
a) that specified from a service point of view between the user and service provider (e.g PSTN quality, conference quality etc) This is the domain of the speech experts and can be characterised by Listener Speech Quaklity (MOS), end to end delay, and absolute category rating, R.
b) application specific parameters, (e.g. equipment delays, codec choice and performance, codec frame size, packetisation arrangements, jitter buffer design, overall packet loss etc.) Optimisation of all these will determine what can be delivered in a).
c) transport parameters for a given choice of application parameters. This boils down only to three parameters as far as I cna see: tranport network delay, packet delay variation in the transport network and packet loss in the transport network. Again these parameters will determine the results in a) for a given choice of the parameters in b). These parameters are generic from the perspective of the transport network. i.e the transport network does not need to know the details of the application.
So the sequence of cause and effect and control is:
a) User requests QoS class from service provider, b) Service provider determines application specific parameters in conjunction with users equipment and other service providers, c) Service provider requests required delay, delay variation and packet loss from network provider.
I see no need for mapping here. The only QoS info flows within the application are specific to the application and those between the application (service provider) and the transport network are generic. i.e. delay, jitter and packet loss. Have I missed something?
2. The issue of bit rate and media stream statistics I think need to be decoupled from QoS. These are specified to enable optimisation of resources within the transport network. They have no QoS significance from an application point of view. i.e the apllication does not care about the media stream bit rate and statistics but the transport network provider does as it eats up his resource. They may be used for policy enforcement however in the transport network so they do need to be agreed between service provider and network operator. i.e the network operator agrees to provide a given QoS level (delay, jitter, packet loss) provided the media properties are within an agreed profile (bit rate, flow statistics).
3. The next point is how can the service provider know the statistics of a particular VBR stream? These can only be specified over a large number of similar calls and will depend, for instance, on who is speaking, the nature of the speech interaction etc etc. They can only be measured not calculated. The service provider is in no better position to measure these than the transport network operator and, in fact, where no gateways are involved, may not be able to. On the other hand the class of signal would have to be signalled to the network operator for him to be able to distinguish which class a particular measurement belonged to. e.g voice/speech/data, codec type, conference, multicast etc. So I see no purpose in trying to exchange statistics between the service provider (application) and transport operator. I think peak bit rate is all that can be meaningfully excanged. The specification of media class is however perhaps worth exploring.
4. The controlled category has always puzzled me. I only see two possibilities. Either the requested QoS level is guaranteed (on a statistical basis e.g 95% of all connections over a specified period) or not guaranteed. Is your controlled category a way of saying guaranteed, not to 95% but to some lower figure? If you can't put a percentage on it then it seems it is plain and simple not guaranteed. Anything that is not guaranteed to some specified statistical level is best effort and you can't say anything more about it. So I only see two categories here.
In summary, I think we need to do three things in Annex N.
a) Figure out the QoS information to be exchanged within the Application between service providers and end users. This will go in H.225.0 and H.245.
b) Figure out how we are going to signal QoS and media information between the application (service providers) and transport domains (IP or ATM networks etc). The info is basically delay, jitter, packet loss requirements and peak bit rate. We need a protocol for this.
c) we need to work out the interactions between the application QoS signal flows and the application/transport signal flows. I don't think we need worry about how the transport network mechanisms assure the requested QoS paramerters. RSVP/Intserv, Diffserv, MPLS, ATM, over provisioning are all possibilities.
Would welcome comments and views on the above.
Mike
Mike Buckley +44-1457-877718 (T) +44-1457-877721 (F) mikebuckley@44comms.com
----- Original Message ----- From: "Roy, Radhika R, ALCOO" rrroy@ATT.COM To: ITU-SG16@MAILBAG.INTEL.COM Sent: Thursday, August 10, 2000 10:15 PM Subject: H.323 QOS
Hi, Mike and All:
It is time to discuss about H.323 QOS.
I believe that we have an agreement as follows:
· H.323 QOS MUST be backward compatible to support RSVP and ATM QOS as it exists for H.323v2/v3/v4 · Like H.323 spec, the application level H.323 QOS MUST be independent of the transport layer QOS and should support all transport networks (e.g., IP, ATM) · A mapping table between the H.323 QOS and the transport layer QOS (e.g., IP QOS [DiffServ, RSVP, etc.], ATM QOS [CBR, rt-VBR, nrt-VBR, ABR, etc.]) should be provided.
From the H.323 multimedia application point of view, there are following
performance parameters can be used to characterize the traffic characteristics:
· Bitrate characteristics: Peak bit rate (PBR) or peak rate (PR), Sustained bit rate (SBR) or average rate (AR), minimum bit rate (MBR) or minimum rate (MR), and mean bust size (MBS) · Delay and loss characteristics: end-to-end delay (EED) or delay, end-to-end delay variation (EEDV) or delay variation (DV), and bit-error-rate (BER) or (packet) loss rate (LR)
We can now form a table with all parameters as follows:
Table 1: H.323 Multimedia Application Performance Matrix Audio (codecs)--- Video (codecs)--- Data (T.120) PBR/PR Yes/No/Value Yes/No/Value Yes/No/value SBR/AR Yes/No/Value Yes/No/Value Yes/No/value MBR/MR Yes/No/Value Yes/No/Value Yes/No/value MBS Yes/No/Value Yes/No/Value Yes/No/value EED/Delay Yes/No/Value Yes/No/Value Yes/No/value EEDV/DV Yes/No/Value Yes/No/Value Yes/No/value BER/LR Yes/No/Value Yes/No/Value Yes/No/value
From the above table we will have the opportunity to choose each parameter
for each medium (audio, video, data) that makes sense from the application's and enduser's point of view. Again, these parameters can be specified as follows:
· Guaranteed: The value specified for each parameter MUST be guaranteed. · Controlled: The value specified for each parameter MAY be satisfied as far as practicable (possibly with certain range), but definitely NOT guaranteed. · Best effort: No commitment will be made.
Now each medium (e.g., audio, video, or data) will have different categories of performance matrix depending on its selection criteria and this can also be mapped to RSVP, ATM QOS, and others, if needed.
Once we agree on this format, the next step is to create H.323 QOS signaling messages.
This is my input for discussion in the upcoming Portland Q.13 meeting for H.323 QOS.
I like to see the comments from other members as well.
Best regards, Radhika R. Roy AT&T +1 732 420 1580 rrroy@att.com
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ For help on this mail list, send "HELP ITU-SG16" in a message to listserv@mailbag.intel.com
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ For help on this mail list, send "HELP ITU-SG16" in a message to listserv@mailbag.intel.com