Re: [xrblock] WGLC - draft-ietf-xrblock-rtcp-xr-jb-02.txt

[Qin Wu <[e-mail]>]

----- Original Message ----- 
From: "Glen Zorn" <[e-mail]>
To: "Qin Wu" <[e-mail]>
Cc: "Glen Zorn" <[e-mail]>; "xrblock" <[e-mail]>
Sent: Saturday, December 15, 2012 1:55 PM
Subject: Re: [xrblock] WGLC - draft-ietf-xrblock-rtcp-xr-jb-02.txt


> On 12/14/2012 06:20 PM, Qin Wu wrote:
> 
> ...
>>>
> >> The second sentence says:
> >>
> >> These metrics are used to measure how the jitter buffer behave as a
> >> result of the jitter and applicable to a range of RTP
> >> applications.
> >>
> >> The English language usage needs to be cleaned up a bit,
> >
> > [Qin]: Okay, maybe we can say: "
> >
> > These metrics are used to measure how the jitter buffer behaves
> > according to a result of the
> >
> > Packet Delay Variation and can be applicable to a range of RTP
> > applications. "
> >> but more importantly the statement sees to contradict the first
> >> sentence of Section 3 which says
> >>
> >> This block describes the configuration and operating parameters of
> >> the jitter buffer in the receiver of the RTP end system or RTP
> >> mixer which sends the report.
> >>
> >> How can both be correct?
> > [Qin]: I don't see they contradict. The configuration parameters are
> > referred to Jitter Buffer Configuration set in the block header while
> > the operating parameters are referred to those metrics included in
> > the payload of the block.
> 
> OK, I see; that was not at all obvious to me from the current text
> 
>>
> >>
> >> Sorry, but I have no idea what the third sentence means.
> > [Qin]:What the third sentence means is usually we don't consider how
> > the terminal affect
> >
> > real-timeapplication quality. however in pratice user may have
> > different experience
> >
> > to the application quality when they choose different terminal to
> > playout the received RTP stream,e.g., ipad,PC,smartphone with
> >
> > difference screen size, resolution and different buffer size to
> > process the data.
> >
> > So maybe we should rephase the paragraph in the section 4 as
> > follows:
> >
> > "
> >
> > Real-time applications employ a jitter buffer to smooth out jitter
> >
> > introduced on the path from source to destination. These metrics
> >
> > are used to measure how the jitter buffer at the receiving end of
> > RTP stream behaves according to a result of the
> >
> > Packet Delay Variation in the network and applicable to a range of
> > RTP applications.
> >
> > These metrics reflect how terminal related factor affects real-time
> >
> > application quality and useful to provide more accurate end user
> > quality experience.
> >
> > "
> 
> I have a different suggestion, now that I understand what you're trying 
> to say :-).  How about this:
> 
> Real-time applications employ a jitter buffer to smooth out jitter 
> introduced on the path from source to destination.  These metrics are 
> used to report how the jitter buffer at the receiving end of RTP stream 
> behaves as a result of jitter in the network and are applicable to a 
> range of RTP applications.
> 
> These metrics reflect how terminal-related factors effect real-time 
> application quality and are useful to provide better end-user quality of 
> experience (QoE).

[Qin]: Your proposed change looks good, thanks.
> ...
> 
>>
> >> *Section 3.2* The description of the "jitter buffer nominal delay"
> >> says that "It is calculated based on the time for packets spent in
> >> the jitter buffer." How is it calculated?
> > [Qin]:It will be calculate based on the difference between the time
> > when the packet enter into the jitter buffer and the time when the
> > packet come out from jitter buffer. It is mostly up to specific
> > implementation and doesn't need to spell out.
> 
> So it's OK if two different implementations produce different numbers 
> from the same data?  That doesn't seem very interoperable, somehow.  
> Maybe one problem I'm having with this is that the way the word 
> "nominal" is used doesn't map to any definition of which I'm aware.

[Qin]: Suppose a lot of packets are receieved in one measurement interval. Some packets arrive very earlier, some packets arrive very late.
some packets arrive extractly on time. Nominal packet are referred to the packet that arrives exactly on time
Since they have the same reference measurement point, usually they will get the same result.

For the definition, you also can see the metric "jitter buffer nominal delay  " defined in RFC3611,section4.7.7.
> ...
>