SoC (Sip Overload Control)
WebEx Interim - I
Tuesday, Jun-29-2010 at 11:00 EDT (17:00 CET; 15:00 UTC) for 1.5hour
Chairs: Volker Hilt& Salvatore Loreto

(many thanks to Paul Kyzavit and Vijay Gurbani for taking notes)
(attends list at the end)


Reason of the meeting:
-----------------------
Try to speed up the work in the newly created WG. Also,
for Maastricht we want to focus on the mechanism itself. So
this virtual meeting is to synchronize on all the work that
has gone into Overload Controlo (oc) while it was at sipping/dispatch.


Agenda presented and accepted.
- 11:00-11:10 Welcome to the WG, Agenda Bash, Logistics (Chairs)
- 11:10-11:40 Summary of Design Team (Eric Noel)
- 11:40-12:00 Design Consideration Draft (Volker Hilt)



Summary of Design Team, Eric Noel
----------------------
IETF chartered a design team on oc; four independent simulation
models were developed and calliberated against each other.

Presented benchmark method -- 2 core proxies, 5 edge proxies.
R-R used, 7 message call setup.

Went through the oc control algorithms of various teams.
ATT used rate- and window-based oc algorithms.
Sonus used sender- and receiver-based flow controls.
Columbia used window-based oc algorithm.
Bell Labs/ALU used loss-based oc algorithm.

There was question about references. They are in these slices and also
in the overload control design draft.

OC algorithms are stable even in face of large retx delays (slide 11).

Results presented for UDP transport only. Some papers include
studies on TCP, but the most work focuses on UDP.

Paul: Did you only use proxies or some other servers (SBCs, gateways).
Not clear how these results extend to non-proxies.
Eric: Focus on server-to-server control only.

SBCs and GWs also require overload control, it is a real problem
that needs to be investigated to check eventual differences with proxies,
but is out of scope for now (many people agreed).

???: Resources that we consider are those that are required to
process a SIP message not taking into considerations external resources like
codecs, DS0, etc.

Volker: We are trying to achieve SIP servers to handle incoming loads and
this is a different problem than a SIP server running out of audio resources.

Janet Gunn: We want to make sure that signaling survives 9/11 type of
emergencies.

Some discussion on whether an overloaded proxy should give preference
to in-dialog requests. Generally, yes. More specifics on the list.

Question if design team is planning to do more work.
Volker said no more planned, but if people would like more they could
revive the team and the simulators.


Overload design, Volker Hilt
----------------
Salvatore: Design draft to be sent to IESG by Sept. 2010. Please
send comments on it. After Maastricht IETF we'd go to WGLC.

This draft does not define a solution, just documents the design
space. Solutions relegated to individual drafts. More on these
in Maastricht.

Discussion on implicit and explicit overload control.

Discussions on Local Overload Control slide. General feeling seems
to be that it is a good thing to have as a last resort. Nothing to
standardize and can be used without any extensions.

Eric: But it is not sufficient to prevent congestion collapse.

Some more discussion on whether SIP extensions needed for implicit
overload control. No time for detailed discussions, asked to be
taken to the list.

Went through rate-based and loss-based oc, window-based oc and
signal-based oc (503).


Discussed the process by which an upstream node selects messages to be
redirected/rejected to reduce load.
Paul: Should we leave priority to implementations or provide more
guidance?
Discussion ensued, but no consensus. Need to take it to the mailing
list.
Paul: what time window should be applied for computing the rate?
No answer.

Meeting ended @ 11:38a US Central (10m over schedule).

Attendees:
- Nick Stewart, ATT
- Salvatore Loreto
- Eric Noel
- Ahmed Abdelal
- Anant Ghotkar
- Carolyn Johnson
- Christrian Schmidt
- Geof Hunt
- Janet Gunn
- Mary Barnes
- Muthu Perumal
- Padma
- Partha R
- Paul Jones
- Paul Kyzivat
- Richard Kaczmarek
- Robert Sparks
- Sheshadri Shalya
- Victor Pascual
- Vijay Gurbani
- Volker Hilt