Notes, SIPPING Session 1, IETF 56
Reported by Dean Willis
Scribes: Vijay Gurbani, Mary Barnes
Chat coordinator: Al with the Hat
Topic: Service Examples, Alan Johnston
Moving well. No comments.
Topic: cc-transfer, Alan Johnston
Ready for Last Call proposed, no objections.
Topic: Status report: Chairs
Note chairs believe that formal expert review is not required for all
individual drafts. Current policy is to announce the draft to the list,
and ask the participantst to verify that the draft does not conflict
with WG efforts.
Topic: Caller-Preferences Use Cases, Jonathan Rosenberg
Poll: Useful, interest in accepting as WG document? Strong consensus.
Suggestion from AD that use cases be packaged with CallerPrefs drafts.
Topic: Persistent Connections, Vijay Gurbani
Basics presented. Discussion ensued. Question: What is the requirement
difference between signaled persistence and simply configuring
persistence? Assertion: Reuse is only important in the client-server
case. Assertion: This sounds like my kids discussing bedtime. There is
really only reason to close a TCP connection -- to recover resources.
Therefore, the interesting thing, is not how do connections come up,
but figuring out which connection to kill if we need to kill one.
Assertion made that the mechanism here is equivalent to "just doing the
right thing" with TCP, so we should fix it operationally, not by
changing the protocol. Hum: Should we change the SIP spec to encourage
leaving TCP connections open unless they need to be closed? Strong
consensus reported. Proposed that we do minimal bug fixes in SIP, and
do a separate document with an operational focus on "how to" reuse TCP,
initially as a discussion and possibly leading into a BCP. No
objections noted, Vijay Gurbani may volunteer to lead on usage
document.
Topic: ICE, Jonathan Rosenberg
Provides general methodology for using address-fixing protocols to get
SIP through NAT. Question: What is the impact if some sort of
underlying mobility function causes node address changes? Does ICE
resolve? Ans: Unknown, but we think it should work if the frequency of
address changes is lower than 1/RTT. Question: Does having STUN server
on every node open up the network to denial-of-service attacks on NAT
translation? Ans: No -- at least, no more than port-scanning already
would. Poll: Who read: fairly large number. Question: Muxing STUN and
media -- is that required? What is the implication of demux? Ans: It
seems to be feasible to use magic-cookies in the TUN level to work it
out. Comment: This DOES require widespread deployment to be useful, can
we get that? Ans: If it works, people will implement it. Question: What
are implications if only one end supports? Ans: We fall back to
what we have now. Question: IS it easier to fix the NAT than to fix the
SIP? Ans. Debatable. There's a lot of $39 NAT boxes out there.
Poll: How many people would be interested in implementing? (several)
Will one volunteer to write the preconditions (document things like
like SDP changes, requirements in SIPPING) Yes-- Cullen Jennings. Poll:
Would you like this work to be documented in SIPPING, along with
fallback to previous modes? Ans: Loud response, a few dissenters.
Proposal: Cullen to work on preconditions, Jonathan to work on
furthering ICE draft
Topic: Session Policy Requirements, Jonathan Rosenberg
Reviewed scenarios. Poll: Do we want to work on this? Brian Rosen
reports that he really needs this for dealing with 6Mb/s streams and
managing admissions policy dynamically. Another comment made that this
approach could obviate the need for some B2BUAs. Comment that we should
add requirements to do this without "new headers" and it shoud work
with e2e encryption (req #1 in presentation). Poll: who thinks we
should take this work on as a baseline for requirements? Strong
consensus reported, none opposed.
Topic: URI Leasing, Jonathan Rosenberg
Material introduced from slides. Question: This allows for temporally
scoped URIs. Can we give an example of where this would be useful? Ans.
Assume a conference call running over a set time. Discussion that
current document may be a bit complex in this area. The word "temporal"
is probably misapplied in this context and the author would like to
withdraw it. This all comes down to embedded route headers vs. a
lease-and-preference model. Perhaps we need to step up a level and ask
what the top-level requirements really are? The draft proposes three
explicit. There may be also "without hiding state in the domain name"
and some others. Comment: Leasing seems to imply temporal scoping. Huh?
Long discussion . . . Comment: It seems that all of the given
requirements could be met by relaxing a single restriction in
3261. Conclusion: More list discussion on requirements needed.
Topic: App Interaction Design Team, Jonathan Rosenberg
Minimal activity since last meeting. We may need to consider use cases
illustrating the requirements.
Topic: Conferencing Design Team, Alan Johnstson
Requirements, Framework, and Call Control ready for WG adoption.
Conference Policy is splitting into two and should be ready shortly.
Media policies work requirements progressing, work on policy
manipulation is ongoing, and the policy work needs a new home. MMUSIC
is not going to adopt this in their new charter. Observation: This
policy work is related to data manipulation in SIMPLE, can we combine
it? One is data manipulation, the other is policy RPC.
Open issues recapped in slides. Discussion of absolute vs. relative
time in focus. Why not have absolute time? Ans. Simplifies timeing
coordination on focus. Comment: We do have requirements for realtime
transitional controls for things like camera-follows-speaker. SOAP and
ACAP don't combine to meet this requirement. Ans: This is for more
persistent policy, not floor-control. Floor-control remains a media
problem.
Poll: Do we wish to, as proposed, adopt the requirements, framework,
and call control drafts as wg efforts? Strong consensus, no objections.
Question: Is discovery in-scope? It could be a furball.
Request: Make focus migration happen sooner rather than later. This
will be required for three-way to more-way migrations. This is also
important from a fault recovery standpoint. The call-control is easy,
just a REFER series, but handling the policy migration is more
interesting.
Topic: Hearing impaired media transcoding, Gonzalo Camarillo
Work proceeding, no surprises.
Topic: Emergency calling design team, Jon Peterson
Work proceeding, will have face-to-face meeting here. Two cases --
regular 911 calls, and PSAP access-link replacement. Can we divide
these two cases so that there is something available for "today's PSTN
PSAP"?
Topic: Event Throttling Requirements, Aki Niemi
Notification rate limiting is the "common factor" among most of the
event filtering discussions. Issues: Is the model accurate and
appropriate? Do we need both the leaky bucket and the strict throttle?
Is it ok to leave handling of quarantined notifications out of scope?
Is this work useful?
Discussion: It may be important to have a different throttling
mechanism for "aggregated/filtered" notifications than for simply
prioritized or decimated notifications. The current work does not
really consider notification volume/size, just counts. Discussion about
the difference between quarantine and gating/throttling functions. One
can define a five-token multibucket scenario that's fairly complete. Is
this too complicated to be practical? Volutneers to review reqs --
Jonathan, David Oran, and Tomn Taylor volunttered. Poll for adoption:
none opposed. Question: Is there standardization required to be able to
rate limit?