[asterisk-users] T.38 - Which endpoint shall reINVITE ? caller or callee ?

[Klaus Darilion]

Steve Underwood schrieb:
> Hi Olivier,
> 
> Olivier wrote:
>>     T.38 says that if the call starts in audio mode it is the called end
>>     which should initiate a re-invite to change from audio to T.38. This
>>     makes sense, as that is the end which has the best chance of figuring
>>     out if a FAX machine answers the call. In practice many T.38
>>     implementations will send out a re-invite when they are the calling
>>     side, so any practical implementation has to allow for this.
>>     Clashes are
>>     possible, if both ends send re-invite, and this is not always handled
>>     properly
>>
>>
>> Yesterday, with 2 consecutive sendings on the same setup (same fax 
>> file, same ATAs, same servers), on the first try, I've seen the 
>> reINVITE coming from callee on from the caller on the second try.
>> I don't remember I changed anything between both tries (though I may 
>> have done without noticing this).
> That is what typically what happens when the calling end doesn't obey 
> the spec. It comes down to a race for who initiates the re-invite first. 
> If you are lucky the two ends sort themselves out. If you are unlucky 
> you end up with both ends re-inviting, and you may get a call failure.

This is what I see very often - both sides send reINVITE (overlapping), 
both sides reject with 491 (request pending) with the result that the 
switch to T.38 failed. I have only once seen handling overlapping 
reINVITEs smart (ComISDN client)

Theoretically it would be perfect if the reINVITE is always triggered by 
the callee (as the spec says) - but there are ATAs which need up to 15 
seconds to detect a fax and send reINVITE - and in this cases you 
sometimes have to work around by allowing the caller to send reINVITE too.

regards
Klaus