Notes, SIP WG, IETF 56
reported by Dean Willis
Scribe: Ramu
Chart Room Coordinator: Not recorded
Topic: Non-invite transactions, Robert Sparks:
Poll == who as read and understands the draft, app. 20%. Principle
problem is the race condition of 64*T1 built into the protocol. Given
non-zero latency, the UAS side has an offset on this time, so the
client may time out before the server completes. Now we just try and
"complete as soon as possible" giving response codes like 408 that make
no sense -- by the time it gets back to the UAC, the UAC is assured of
having time out, crating a 408 storm. Other problems exist with
intermediary timeouts and with forking convergence.
Two solutions offered, A and B. Rohan argues against proposal B,
preferring the subscribe/notify style of seperate transactions. Point:
What if values of T1 are different? Discussion: If UAS and UAC have
different values of T1, it's all broken anyhow. Extended discussion
follows. Two problems: the infrequent 1 in 10,000 events, and the
problem that the 200 has to be emitted soon enough to have a high
probability to make it back to the UAC before the 408s. Most troubling
is when the UAS thinks the situation is 200ok, but the UAC has a
different view (which can happen with non-invite forking). Point that
we should lift the constraint on interoperability with existing
proxies. Suggestion we adhoc during this meeting for consensus -- group
to email Robert for gathering.
Referred-by, Robert Sparks
Complain: nobody is giving feedback. Are people mostly happy and just
waiting on some of the blocking work to complete? Volunteers to provide
feedback: Brian Rosen, Mary Barnes, Alan Johnston, Pekka Pessi.
Issue: Use of Call-ID in auth-ID-body. Is it a leak/ Should we put
garbage in, change the AID body format, or what? Jon Peterson reports
that the newest AIB draft relieves this requirement.
Issue: Referred identity binding between REFER and INVITE. Can a
referee assert a different identity on the generated INVITE than was
referenced in the REFER? Can humans make the distinction?
Resource-Priority, Henning Schulzrinne
Goal: Higher priority of emergency call completion during service
disruption of civil emergency, especially in interworking with PSTN.
Requirements are establised in RFC3487. List discussion reviewd,
best option a new header. Discussion about whether it is "good" to have
proxies understand this. Consensus: Work on this problem, starting from
this draft as a baseline, but address additional opinions and
requirements as raised in the WG process.
Caller Preferences, Jonathan Rosenberg
Volunteers to review -- Robert Sparks, Mary Barnes, Pekka Pessi,
Bob Penfield, Cullen Jennings, Plan to revew, feed back, revise,
go to 2 week WGLC.
Congestion Safety
Willis - Presented. Nobody understands how it works. People have the
vague feeling its important. Khartabil - Implementors do not see
immediate value. General belief that we should consider reposnse-CI as
a possible alternate mechaniusm in primary work.
Auth-ID Issues, Jon Peterson
The only new "SIP" function is the new 400-response code. Other
functions are non-normative.,