TSVWG Session (sent to the list April 7th, at 2pm for corrections) Wednesday, March 22nd, 2006 0900-1130 Attendance 36+19=55 people Chair's 9:00-9:15 Agenda Bashing (15 min) NOTE WELL Document Status and Accomplishments New Charter? New Milestones? No Agenda Bashing - Matt said his TCP MIB Ext doc is ready for WGLC (see below) - During Doc Status - to consider draft-ietf-tsvwg-vpn-signaled-preemption WGLC worthy (hum 100% for, 0% against) Bar BoF formal WG Announcement - Gerald Ash was concerned of the scope of the bar BOF. Not prepared to present work there. - Georgios balked at not having time to prepare slides - Allison said to deal with it Aaron's 9:15-9:25 (until 9:26) Congestion Control Research Group (10 min) - Discussed congestion control RG - Problem: High throughput requires unrealistic low packet loss rate, ==> low throughput in lossy environments - Plan 4 Phases Challenge, Solution Proposals, Discussion, Consensus; - Rec to IETF. Michael and Randy's 9:25-9:35 (until 9:39) draft-tuexen-tsvwg-sctp-padding-00 (10 min) (might want to add SCTP auth and addip to here) - Padding chuck is needed by path MTU discovery - Need motivation in doc - Lars suggested that the doc be split in two (padding and the other stuff) - Matt said a lack of motivation is good to stay as is - Michael said - Hum to adopt as WG item (knowing sec sect needs a bit more) - Hum 100% for adopting, 0% against - ADDIP - Not intended for mobility, unlike DCCP - Need to slightly modify before progressing to WGLC - SCTP-AUTH has known dependencies that are needed - AUTH Only want to secure ADDIP - Needs clarification of Null key, Key Identifier - Need to slightly modify before progressing to WGLC - Must go together with ADDIP through process Kwok's 9:35-9:45 (until 9:51) draft-chan-tsvwg-diffserv-class-aggr-03 (10 min) - Fred presented - Motivation to provide less granular treatments - Editorial changes - Changed name of default class - Dave McDyson believe this should go to WG status and his network will use this - Dave Blacks thinks this is a good idea and should be a WG item - Hum for WG status (100% pass, 0% against) - Dave Black will be a reviewer - Magnus thinks this ought to have mail-list confirmation for WG status Phil's 9:45-9:57 (until 9:57) draft-briscoe-tsvwg-cl-phb-01.txt (12 min) draft-briscoe-tsvwg-cl-architecture-02.txt - Split document into two (a deployment ID and a marking ID) - CL-Arch has changes (on a slide) - CL-PHB has changes (on a slide) - Problem: they have 5 things to encode, and 4 possibilities, so creativity is necessary - Asked that folks come to the Bar BoF and give comments and feedback Bob's 9:57-10:15 (until 10:21) draft-briscoe-tsvwg-re-ecn-tcp-01.txt (18 min) draft-briscoe-tsvwg-re-ecn-border-cheat-00.txt - RE-ECN-TCP - About adding accountability to IP congestion - How to allow some networks to do policing. - Allow different networks to deploy different policies some conservative, some liberal. - Backwards compatible with the Evil Bit - Goal is also to motivate the use of ECN to allow congestion marks to be able to be taken advantage of by SPs - Question from someone - Problem finding the extra bit in a IPv6 header, but there is the problem of by adding a bit - you need to add a whole 128 bit extension - Dave Black wants to open up discussion on using the last bit as a real discussion point - He's also concerned about the differences between this solution in v4 vs. v6 (which isn't public yet) - Allison offered that perhaps a creative use of the flow label in v6 could be the answer - RE-ECN-Border - Chairs only allowed 1 slide of preso due to time - Gave short summary only Matt's 10:15-10:25 (until 10:24) draft-ietf-tsvwg-tcp-mib-extension-08.txt (10 min) - MIB Dr didn't give many comments - MIB Dr said it's basically done - There was a template problem that wasn't caught, but this can be corrected during the WGLC process - Ready for WGLC Francois's 10:25-10:40 (until 10:53) draft-lefaucheur-emergency-rsvp-00 (15 min) draft-ietf-tsvwg-rsvp-ipsec - RSVP-IPSec - Removed SPI from what identifies the Reservations - This made everything here simpler - Added notion of extended virtual destination port, also used in MPLS TE - Hum to WGLC (100% pass) - Agenda had a mistake that this was a individual status (the chairs take the hit on that one) - RSVP-Emergency - Discussed capacity handling in a number of ways based on a local configuration - Draft presents RSVP extensions to identify Emergency sessions and give them elevated probability of session establishment - Speaker asked to take this - Gerald Ash - how is this mechanism mapped to the RPH? - James answered - that SIP is disconnected from Router behaviors, and that local policy determines the mapping from the RPH to RSVP Priority values and treatments (Note that the draft now proposes an extension to also convey the RPH inside RSVP signaling) - Janet commented - - Georgios asked, in the case of the new Priority Bypass Model, about if there are too many higher priority calls, at what point, at what capacity does this doc suggest to stop admitting new emergency calls or terminate non-emergency calls - Francois answered that, with the Priority Bypass Model, new emergency calls are always accepted and non-emergency calls do not get preempted. It is a simple model that works under the assumption that emergency load remains reasonable. The other models described in the draft can be used to deal with other cases (e.g. stop admitting emergency calls at some stage). - Allison commented on reorganizing the doc by separating the BW model from the mechanism, as this ID is clearly a PS track but the model is not - so it can be in an appendix - Francois agreed. This will be reflected in next rev. Sally's 10:40-10:50 (until 11:09) draft-ietf-tsvwg-quickstart (10 min) - Mitchell Airblix (sp?) comments off-line will cause a few more paragraphs to be added to the ID - Tim Sheppard - Multi-Access link Ethernet scenario is clearly problematic (and Sally agreed) - This won't work if TCP header is encrypted (military nets) but there are some nets where this will work. - Assumption is that this mechanism will never ever be a default enabled feature - Fred - asked who the audience to be then? - Sally - expects to be used in very high BW little used intra- networks, ID clearly lists numerous cases in which this mechanism will not work - including anytime a packet in tunneled - Bob Briscoe - believes this could work in a server farm - Randy Stewart - asked about IPv6 - Sally - this is fine in IPv6 - Lou Berger - believes this is limited in scope Jukka's and Markku's 10:50-11:00 (until 11:19) draft-manner-tsvwg-lrsvp-00 (10 min) - This is not a new ID - Motivation: clearly e2e QoS is not here, and in times that the remote node is not RSVP enable, this mechanism allows RSVP as far as it can, and not mandate e2e or even a mechanism at the remote node. - Reviewed the 2 extensions necessary to enable this mechanism - Wants 2 bits from Session Object, 2 new message types. - Francois - thinks there is value is documenting proxy behavior, not clear that the endsystem needs to be the starter of this mechanism (i.e. a L-RSVP proxy in a early router, more like edge-to-edge) - Will send comments to the list - Bob - interested how a pair of hosts will ever know if there are in this situation or not, perhaps adding a new protocol ID could help identify it - Will also send comments to the list now that he understand this (from the preso, as he didn't from the ID itself) - No vote for WG status vs. the slide suggestion of moving to an EXP status - Chairs asked for list comments before going forward in either direction Jonathan's 11:00-11:10 draft-rosenberg-sipping-overload-reqs-00 (10 min) - This preso was skipped Transport Area Chairs Update 11:10-11:25 (until 11:32) DCCP Chair (Lars Eggert) (15 min) - Gave overview of DCCP to TSVWG - specs pending. - WG being rechartered. - New chair Gorry RMT rep (Allison stepping in) - Reliable Multicast WG is moving their documents from experimental to proposed this spring. Flute/ACL with a NACK protocol. One called NORM with building blocks.. giant group of protocols going to PS.. must look at very closely... IPPM rep (Matthew J Zekauskas) - Gave overview of IPPM to TSVWG - Chartered to do metrics for paths - list of them on slides. - Cooperate with ITU - Did all the low hanging fruit first (slide 7) - No bulk transfer capacity metric yet. - Slide 8/9 more completed work. - Still working .. stuff in page 10. reordering moving forward WGLC. ==================================== Review of Hum votes - draft-ietf-tsvwg-vpn-signaled-preemption-00 considered WGLC worthy (hum 100% for, 0% against) - draft-ietf-tsvwg-rsvp-ipsec-00 considered WGLC worthy (hum 100% for, 0% against) - draft-tuexen-tsvwg-sctp-padding-00 to be WG item (hum 100% for, 0% against) - draft-chan-tsvwg-diffserv-class-aggr-03 to be WG item (hum 100% for, 0% against) ==================================== CL+RMD Bar BoF attendance: 13+15+5=33 Chairing James Polk Francois on the overall approaches. Edge2Edge/Controlled Deployment Model David Black: Assumption that all in PCN domain are in on the PCN. Inside the domain there is no interaction between PCN and ECN. PCN is associated with some changes in the ECN field treatment. What are opportunities if there are mixed ECN and PCN end to end. The model discussed is an Edge to Edge/controlled. Important to consider these models when doing the discussion. See first model slide for this model. The end system does not need to be trusted. There is the use of normal RSVP to the end system. End to end/open deployment model is the other version. Consider mixed usage model with PCN and ECN. Questions: A) Focus on Edge2Edge/controlled? B) Decide that we want to address both Edge2Edge/Controlled and End2End/Open? c) Work out viability of End2End/open first? Are there other models to cover? Pratik Bose: Will ECN have the same meaning in end to end and edge-to-edge? ?: Experimental solution for end to end. Use different marking system and how would they be used? Need input how fast we should go forward. David Black: Focus on A. Mechanism are used seem to be fine granularity adjustments that would not be available in end-to-end. Figure out co-existence with normal ECN (RFC 3168). Bob Briscoe: Started thinking about the end-to-end case and could not easily resolve it. That why we started with edge to edge. Requires reasonable behavior from systems, more likely in network than end hosts. Encoding -------- Presentation of 4 alternatives. Bob Briscoe: Do people believe this should be done in ECN field or in the DSCP field? Francois: Why would you not use DSCP in a controlled environment? What is the motivation for this? John Anders: Is okay to use another DSCP codepoint, like alt 1? David Black: It is no obvious that you need to use another code point as you could filter them into another traffic class and not handle that with PCN end to end. Francois: Filter voice into its own DSCP. Have the PCN behavior depend on the DSCP. James: Is this a local decision on what to use? David Black: No, especially for alt 1. One thing that is happening. Not have the ECN field to depend on the DSCP. Bob: Use AM control to traffic that handle ECN. Non ECN is static assignments. Francois: No problem in the controlled environment to use one DSCP. Down path it might be a bigger problem. Like MPLS. EXP is scarce and using EXP for this may not be available Kwok: Pretty open on this issue. Pratik Bose: How about we allow multiple alternatives for encoding and operator decides what they do in their network? David Black: As DSCP are operator dependent this can be a problem. Complexity problem. Would be good to standardize one way to do this. Francois: Good to do this in one way for Hard ware support. Kwok: Not have ECT encoding within a the field. Question on what is better to use. David Black: This is not obvious. Needs to go to the list. Francois: Goal was to get people thinking. Initial simulation Study Anna Charny: Limited applicability. It works in the simulation we have done. Not done a complete verification. This is a proof of concept. And a demonstration that there is ongoing simulation work. PCN/RMD Francois: Pre congestion marking: STD or EXP? John Loughney: Good bring to that into NSIS is supported. There is a individual draft for integrated service control load: Will post the draft ref to the TSVWG list. Bob: Do people think it is enough that PHB and PCN marking are specified independently? Do we need to a top level document declaring that combination that is required to get the service? This would be a one page. Restatement David Black: From process point of view, separating PCN from PHB is a better question. Would provide better clarity. Anna: A PDB document would need to define a lot of behavior and group things together. David Black: It will probably be needed to define how the border handling is done between domains. Jozef: Do we need a PDB that defines both the DSCP and ECN. David black: Sloppy use of PDB. Document needs to define the domain behavior, not necessarily a Diffserv behavior. Bob: Are people here for interest, or to actually contribute. How many people would contribute to this work and review: 14