On Sat, 2009-07-04 at 12:59 -0400, Mark Gertsvolf wrote:
> sipX seems to be configured with rfc3261 T1 timer set to 100msec instead
> of the recommended default of 500msec. 
> Client INVITE transaction generates INVITE request, which is
> retransmitted only 3 times - after 100msec, 200msec and 400msec.
> Then sipXproxy kills the transaction branch only 1.5 seconds after the
> original INVITE was sent. 
> When far end finally responds with Trying and Ringing and possibly with
> 200OK, sipX drops all of these responses of the floor.
> 
> The above means that any outgoing INVITE over UDP, which takes more then
> 1.5 seconds to produce provisional response is doomed. 
> 
> Standard INVITE transaction with T1 of 500msec as prescribed by RFC3261
> section 17.1.1.2 lasts for 32 seconds and should cause 7
> retransmissions. 

We certainly could make the timer and/or retry count values
configurable, but it would be a bit of a pain, since many components
would have to be modified to read the values in and down to the stack.

To be clear - 3261 does not mandate 500 milliseconds.  It says that T1
is an estimate of the round trip time.   The value of 100 milliseconds
we chose is a much more realistic estimate of round trip times on the
modern internet (to say nothing of intranets); at the time we made that
change, we measured round trips well under that between Massachusetts
USA and Bangalore India over paths with hops in the high teens.  

The 6 or 7 retransmissions are, in practice, ridiculous - no one will
wait 32 seconds for a phone to ring; this is a classic example of spec
authors not paying attention to the real world.

> This might be particularly problematic for remote worker deployments,
> where network delays could be significant.

The problem case you've found is really not a network delay problem at
all - it's an application problem.

If we're going to allow for any change to make this better (and I'd
suggest that a closer look at the reason for the delay on the softphone
before we consider doing anything in the stack), I would tune the number
of retries rather than the T1 value.


_______________________________________________
sipx-dev mailing list [email protected]
List Archive: http://list.sipfoundry.org/archive/sipx-dev
Unsubscribe: http://list.sipfoundry.org/mailman/listinfo/sipx-dev
sipXecs IP PBX -- http://www.sipfoundry.org/

Reply via email to