TSV ADs, TSV-Area follks,
Apologies, I missed tsvarea this afternoon (preparing slides, and I
missed my alarm).
I understand the session on current and future hot topics in TSV was
rather short.
Maybe the two sentences above are related ;)
1. I think the DualQ Coupled AQM is an important development, not just
cos we've got cool low queuing delay for all traffic out of it, but...
...because it is an "incrementally deployable clean slate" that is
interesting enough to network operators that it could get deployed.
What I mean is, it's a incrementally deployable queue isolated from
existing traffic in which we could develop new congestion control
together with new network behaviour.
Obviously, new host & network capabilities have to be deployable
independent of the other, but there is a window of deployment
opportunity for new ideas....
For instance, if a new slow-start needs a new signal from the
bottleneck, that could be included in the initial spec for the L4S
queue, so there would be no need to cater for L4S queues without that
signal. Of course, you would have to cater for non-L4S bottlenecks
still, but if you were getting signs you were in an L4S bottleneck, this
could be powerful.
2. The L4S activity itself would involve new TSV standardisation
activity across 3 WGs (not to mention new implementation activity).
We're planning on outlining this in the L4S Bar Bof on Thu 9am
(advertised earlier today on this ML).
Slide 4 of our slides about L4S (tomorrow in tsvwg) shows the three
areas needing standardisation (AQM, TSVWG & TCPM)
http://bobbriscoe.net/presents/1604ietf/1604briscoe-tsvwg-ecn-l4s-id.pdf
There's also potential for loads of new activity evolving scalable TCP
(TCP Prague/DCTCP).
I prepared the following table for the TCPM chairs listing all this:
http://bobbriscoe.net/presents/1604ietf/tcp-prague-todo.pdf
Cheers
Bob
--
________________________________________________________________
Bob Briscoe http://bobbriscoe.net/
