[Iccrg] Congestion Exposure (re-ECN): to BoF or to Bar BoF?
Bob Briscoe
rbriscoe at jungle.bt.co.uk
Thu May 21 09:37:07 BST 2009
TSV Area folks,
[Please continue this discussion on tsvarea]
I'd like to launch a concrete protocol-specification activity on
congestion exposure (with re-ECN as a candidate solution) in
Stockholm? Here at the ICCRG in Tokyo, we've just agreed that this
doesn't conflict with the new IRTF ICCRG design team in this space
or prejudge any consensus on direction (see background below). The
question is purely about timing...
I'm finding that re-ECN seems to feature somewhere in the plans of
those expressing an opinion in the ICCRG discussions. My concern: the
protocol that will take longest needs to be finished soonest. E.g.
ECN took 7 years from first research paper to proposed standard.
Therefore, I'd like to propose we get started on a specific protocol
specification activity, in parallel to the ICCRG design team. We have
to set this up so it doesn't imply it is the *official and only* way
forward. But it's a good punt to be starting on.
My questions are:
* Do you think discussions should start on building a BoF on
"Congestion Exposure" in Stockholm (getting late but still do-able)?
* Or less ambitiously should we arrange another ad hoc BoF (bar BoF)
in Stockholm, building on ICCRG work in Tokyo this week, to get a
community together to form a BoF later?
* Or do people actively oppose anyone doing anything yet on
congestion exposure or re-ECN in the IETF or IRTF?
Background on TCP-unfriendly and on re-ECN
-------------------------------------------
You may be aware that, in San Francisco we had discussions in TSVArea
(prompted by Matt Mathis's talk) on whether to move beyond
TCP-friendliness as a comprehensive direction for both sharing
Internet capacity and future congestion controls. In the IRTF ICCRG
the same week it was decided to set up a design team to work on the
way forward and propose it in an informational RFC - Matt Mathis has
kicked off a first shot <draft-mathis-iccrg-unfriendly-00.txt>. We
just had the launch meeting of that design team at the ICCRG here in Tokyo.
Re-ECN is a change to IP to make congestion as transparent to the
network as it is to the endpoints (hence congestion exposure). Then:
i) causes of excessive congestion can be kept in check if ISPs choose
- but at the bulk packet level like the Internet architecture should
be - agnostic to behaviour of individual flows;
ii) and ISPs that under-provision can be held to account as well.
(see the end for the status of re-ECN).
Re-ECN status
-------------
We've been keeping a draft spinning on this proposed change to IP
since 2005. It's still an individual contribution waiting for the right time.
<draft-briscoe-tsvwg-re-ecn-tcp-07>
<draft-briscoe-tsvwg-re-ecn-tcp-motivation-00>
When we first presented it and had subsequent ad hoc BoFs, the idea
fell on stony ground. I think because back then few could see the
limitations of TCP-friendly. Even if you still think there's
something in TCP-friendly, it was apparent from the lack of any hum
it received in SF that we at least need something wider.
It seems re-ECN is one of those things people want someone else to
have already done. It gives a permit to be able to work on protocols
that are really safe and fair, but they're not strictly TCP-friendly
(e.g. mTCP, relentless, etc). For instance, re-ECN would allow ISPs
to count how little congestion LEDBAT traffic was causing, so they
need not count it against the user's allowance, etc.
We've now implemented it. There's now a great amount of security
analysis on trying to game it and attack it and how it defends
itself. People are working on using it as a basis for other DDoS
protection mechanisms. Another is trying to build traffic engineering
over it. Others are interested in deploying it using proxies. It's
potential impact on net neutrality etc is being analysed by
regulatory and economics people. So has it's time come?
Cheers
Bob
________________________________________________________________
Bob Briscoe, Networks Research Centre, BT Research
More information about the Iccrg
mailing list