Minutes IETF97: tsvwg
Transport Area Working Group
||Minutes IETF97: tsvwg
tsvwg Session 1 (2:00:00)
Tuesday Afternoon session I 13:30-15:30
Room Name: Park Ballroom I
tsvwg Session 2 (1:00:00)
Wednesday Morning session II 11:10-12:10
Room Name: Grand Ballroom I
TSVWG Meeting at IETF 97, Seoul, South Korea --> Minutes [DRAFT]
WG Chairs: Gorry Fairhurst and David Black
Note Taker 1: Randy Stewart
Note Taker 2: Vincent Roca
Jabber Scribe: Tim Chown
---- Session 1 (100 min scheduled out of 120 min total)
2. Status/updates (WG Chairs) (15 minutes)
UDP Usage Guidelines
--> Now in RFC Editor Queue
DiffServ interconnection classes and practice
--> IESG telechat after Seoul, Dec 1
Tunnel Congestion Feedback
--> WG LC Complete, need two more reviews, Bob Briscoe and Jake
Guidelines for DiffServ to IEEE 802.11 Mapping
--> WG Last Call coming in December
Charter & Milestones
--> See slides for milestone updates
New draft: ECN for Tunnels that use shim headers - new
milestone: June (submit for RFC publication) New draft: SCTP
(RFC 4960 Errata and Issues) - new milestone: September (submit
for RFC publication)
--> During the sessions in Seoul, the sense of the room was to adopt
4 more drafts - milestones
will be added for each draft if/when adopted.
3. WG draft-to-be
A Lower Effort Per-Hop Behavior (LE PHB) - David Black for Roland Bless (10
--> Draft now has correct name, is an adopted WG draft
--> David Black (WG chair) suggests that this be standards track, and
RFC 4594 (Diffserv Service Classes), as work on the 802.11
Diffserv draft found some confusing text there that needs
clarification, and also suggests that obsoleting RFC 3662
(Lower Effort PDB) is the right thing to do. Will be taken up
4. L4S - Bob Briscoe (30 minutes)
--> L4S work is expected to take place in TSVWG. Sense of room in
Seoul is that the
WG should adopt these drafts (no hums heard against adoption).
The actual WG adoption call for these drafts on the mailing
list is expected to follow the list adoption call for the ECN
Experimentation draft (see item 5 below).
Low Latency, Low Loss, Scalable Throughput (L4S) Internet Service:
DualQ Coupled AQM for Low Latency, Low Loss and Scalable Throughput
--> Overload options: a) Continue to mark, use tail-drop OR b) Switch
to classic AQM drop
Mirja (AD) suggests b), which preserves low latency when low
latency queue is overloaded
in contrast to a) which fills the queue.
Results shown use drop in non-L4S queue. Have experimented
with ECN there, no significant
difference in results for drop/mark rate less than 25%.
Identifying Modified Explicit Congestion Notification (ECN) Semantics for
Ultra-Low Queuing Delay
--> If there are two or more bottlenecks, and any bottleneck other than
the first is located at
a dual queue node, this marking technique causes packet
reordering for non-L4S traffic at any dual queue node that is a
bottleneck, but not the first bottleneck because all because
all CE-marked packets, including those for non-L4S-traffic, go
through the low-latency L4S queue.
5. ECN-related drafts
Explicit Congestion Notification (ECN) Experimentation - David Black (20
--> Major open issue is whether ECT(1) codepoint usage outside of
experiments should be
strongly discouraged (SHOULD NOT use) or outright prohibited
(MUST NOT use). The rough consensus in the room is for
prohibition (MUST NOT use). This is reflected in the -04
version of the draft uploaded after the meeting, and discussion
of this topic will continue on the list.
--> Sense of the room is for WG adoption of this draft (no hums
heard against adoption).
Adoption will be handled via a formal adoption call on the list.
Guidelines for Adding Congestion Notification to Protocols that Encapsulate
Propagating Explicit Congestion Notification Across IP Tunnel Headers
Separated by a Shim
draft-briscoe-tsvwg-rfc6040bis (NOW draft-ietf-tsvwg-rfc6040update-shim)
--> Quick status update on both drafts. rfc6040bis draft has now
renamed and is a WG draft.
6. Congestion control drafts (brief introductions only, additional discussion of
one or both drafts may take place in the immediately-following ICCRG
Circuit Breaker Assisted Congestion Control (CBACC) - Jake Holland (5
--> Problem: Malicious node joining multicast sessions can overload
link, degrading service to other nodes. Problem must be solved
in order to successfully deploy inter-AS (across multiple
operators) multicast video service.
--> Only a brief introduction in TSVWG, more discussion occurred in
Layer 3 Quantized Congestion Notification (L3QCN) - Yolanda Yu (5 minutes)
--> This draft received a longer discussion in TSVWG, and hence was
not taken up
in ICCRG. The draft proposes to extend Ethernet backwards
congestion notification (QCN) over IP via encapsulation for
delivery across IP connectivity between Ethernet domains.
Important comments from discussion:
- Draft is currently written for data center leaf/spine
networks only. That's
too narrow a scope for the IETF to consider, but the
draft is applicable to other somewhat controlled
networks, e.g., L3VPN. (David Black, WG chair, speaking
as an individual)
- OTOH, backwards congestion notification outside of a
transport protocol session
is not safe for the Internet in general because it's
too easy for an attacker to spoof the congestion
notification. (Bob Briscoe)
- Ethernet backwards congestion notification (QCN) is not
today, hence extending it over IP may not be useful
(Pat Thaler, IEEE 802 First Vice Chair, speaking as an
-- Session 2 (45 minutes scheduled out of 60 total)
Note Taker 1: Maxim Proshin
Note Taker 2: Vincent Roca
Jabber Scribe: Tim Chown
Agenda recap (5 minutes)
7. SCTP-related drafts
Stream Schedulers and User Message Interleaving for the Stream Control
--> Draft is complete, and will go to Working Group Last Call shortly
Stream Control Transmission Protocol (SCTP) Network Address Translation
--> Now that the -ndata draft is complete, expect progress on this
draft before next
IETF meeting in Chicago. WG chairs would like to be able
to hold WG Last Call on this draft before Chicago meeting week.
Additional Considerations for UDP Encapsulation of SCTP
--> This is next in line after SCTP NAT draft above. Hope to have
an RFC 6951bis draft for
the summer IETF meeting in Prague.
RFC 4960 Errata and Issues - Maxim Proshin (10 minutes)
--> Lots of small items, mostly editorial, a few technical. All the
so reviewers are wanted, and the authors are actively
recruiting people to review.
8. Individual drafts
Transport Options for UDP - David Black for Joe Touch (10 minutes)
--> Only two peole in Seoul have read this draft, so no adoption hum
taken. --> Use of this mechanism to realize UDP-Lite with UDP
protocol number could
be interesting if that works in practice.
--> Request to take measurements to figure out how well this works
Internet in general (Lars Eggert). These would be at scale
measurements to see how widely this works (or doesn't), see
Forward Error Correction (FEC) Framework version 2 - Vincent Roca (10
--> Nobody has read this draft.
--> Conclusion from discusion is to seriously rework the draft: just
additional I-D that explains how to **extend** RFC 6363 to add
convolution codes and explain what changes to RFC 6363 are
necessary to enable that. The result should be a much shorter
draft that will be easier to get people to read. There's no
need to reproduce all of the text of RFC 6363 and obsolete RFC.
This would become an additional RFC that will extend RFC 6363,
and will have to be read in conjuction with RFC 6363.
--> Authors have an implementation. An indpendently developed
someone else (or more than one someone else) other than the
authors is highly desirable.