Re: [sipcore] Eric Rescorla's Discuss on draft-ietf-sipcore-sip-push-21: (with DISCUSS and COMMENT) - the COMMENT issues - part 3

[Christer Holmberg <[e-mail]>]

Hi,

S 5.3.2.
>>>   
>>>      If the proxy has knowledge that the UA is awake, and that the UA is
>>>      able to receive the SIP request without first sending a REGISTER
>>>      request, the proxy MAY choose to not request a push notification
>>>      towards the UA (and wait for the associated REGISTER request and 2xx
>>>      response) before it tries to forward the SIP request towards the UA.
>>
>>Why not race these?
>
> One could do that, but I don't think it should be mandated.
>
>> The current text prohibits racing if you *don't* know that the UA is awake.
>
> The basic assumption of the mechanism is that a UA cannot receive SIP requests before it has 
> received the push notification, sent the REGISTER etc.
>
>> But that may or may not be true, depending on whether the device has gone to sleep
>
>> Implementations can of course do whatever they want, but I think it is bad design to specify such 
>> behavior. Unless the request will reach the UA, some re-transmissions will be "wasted", 
>
> We trade off "wasted" transmissions for performance all the time. For instance: RFC 6555.
>
>> and once the proxy receives the REGISTER the re-transmission timer might have become rather big.
>
> Well, that's easy to fix by not incrementing the timer on these transmissions.

That would modify the core SIP re-transmission procedures (and impact the SIP stack). I don't want to do that.

>> But, perhaps we could say something like:
>>
>>"if the proxy has knowledge, or strong reasons to believe, that the UA is awake"
>>
>> An example of such strong reason is that the proxy received the previous REGISTER just a few seconds ago.
>
> This text still seems to discourage this behavior, and I'm not in agreement with you that we should do so.

The reason the default behavior is waiting for the REGISTER request, and the corresponding response, it to make sure that the R-URI is still valid etc. Saying that the proxy can skip those rules whenever it wants, and just forward the request before it has received the REGSTER response, breaks that whole mechanism.

As an optimization, the WG agreed to allow forwarding the request prior to receiving the REGISTER response if the proxy has "knowledge" that the UA is alive. Such knowledge could e.g., be an active security association, an active SIP session, or something else that the proxy keeps track of. If the UA does not have such knowledge, it needs to wait for the REGISTER response. Nothing is broken, so I don't want to change that.

Regards,

Christer