Re: [MMUSIC] WGLC for draft-ietf-mmusic-udptl-dtls-06

[Christer Holmberg <[e-mail]>]

Hi,

>>>> If a new DTLS association is to be established, is it well defined when
>>>> to cut over to it? ISTM that some procedures are required here to ensure
>>>> that data isn't lost, or replicated.
>>>
>>> I don't think we can define any exact procedures for that. DTLS-SRTP doesn't define any exact procedures either. RFC 5763 only says:
>>>
>>>   "Once the new session keys are established, the session can switch to using these and abandon the old keys."
>>>
>>> ...and I don't think we can say much more.
>>>
>>> In addition, I assume there are procedures on the fax application layer for avoiding lost data.
>>>
>>> Also, I think it will be rare that the keys and/or transport parameters change during a fax transmission. Or, if they do, perhaps a fax transmission failure is acceptable?
>>
>> Data loss prevention/recovery mechanisms are in place at the fax layer.
>
> The case I'm talking about is: A new O/A results in establishing a new
> connection, replacing a prior one. At some point during that process
> both the old and new connections are usable. Then, at some point the
> sender stops sending to the old connection and starts sending to the new
> one.

Yes.

> For the receiver to get all the data, it must keep reading from the old
> one until it has received the last packet sent there, and must start
> reading on the new connection soon enough to get the first data sent
> there. And there is a possibility that the packets won't arrive in the
> order sent, so ceasing to read from the new one once something has been
> received from the old one could cause something to be lost. (Or course
> these aren't reliable protocols, so something could be lost in the
> network too.)
>
> Are you saying that the fax layer is sufficient to deal with this? (I
> haven't looked at it.)

Yes.

I don't think the receiver should cease to read from the new one if it receives something from the old one.

If it helps, we could add a sentence saying that the receiver needs to be prepared to receive data on both the new and old connections, as long as they are both alive.

> I realize that this problem isn't unique here - it applies any time O/A
> results in a change to the media path. I've wondered about it for a long
> time, but have assumed it has typically been considered acceptable to
> get a "glitch" in an audio or video stream in such a case. But it is
> less acceptable for fax.

It is not even introduced by this draft, as FoIP using SDP O/A has been around for many years.

Regards,

Christer