IPPM Agenda - IETF 88 Vancouver Notes by Andrew McGregor Thursday 7 November, 13:00 - 15:00 PST (UTC -08:00), Regency A ======================================================================= I. Welcome and Administrivia ------ --------------------------------------------------------------- 13:00 Welcome, Status, and Agenda Bash 10m Chairs (B. Trammell, B. Cerveny) 13:10 draft-ietf-ippm-testplan-rfc2680-04 Last Call Update 5m A. Morton WGLC is finished, comments integrated. Brian Trammell is Document Shepard, he will prepare document write-up and send it to IESG. II. Current Working Group Drafts ------ --------------------------------------------------------------- 13:15 draft-ietf-ippm-rate-problem-02 5m A. Morton Matt Mathis: Have you thought about parameters in terms of registry problems? 2D parameters complicate registries. [15:44] Matt Mathis and Barry Constantine will read and comment. 13:20 draft-ietf-ippm-lmap-path-01 5m A. Morton Matt: Maybe there should be three kinds of link: managed shared, unmanaged shared and dedicated. 13:25 draft-ietf-ippm-2330-update-01 15m J. Fabini Matt: There is science about metrics, and 'useful' decomposes into actionable, meaningful, etc. Near top of Matt's queue. Matt: TCP equilibrium behavior has self-dependence, and doesn't work as measurement. You have to simplify the dependencies. Matt: There are types of measurement only possible for an ISP, which may not be IP metrics at all. Repeatability is what makes for a science. Al Morton: It's a complicated science. Three people said they have read draft and 9 said they will read it by London IETF. 13:40 draft-ietf-ippm-model-based-metrics-01 15m M. Mathis Brian Trammell: There's a lot of timing assumptions here, and the time slotting thing will break them. Matt: Have to talk to Al about that. 13:55 draft-ietf-ippm-ipsec-01 (?) 5m K. Pentikousis Brian Trammell said he would like to hear from TWAMP implementors. [Audio offset: 1:04:00] Al: We?ve split the protocol before, we can do it again if we had to. Kostas: I would not (want to) for many reasons. I can imagine we can make it generic enough that others can benefit later. But we need to hear from TWAMP implementors. [Audio offset: 1:06:00] III. Registry Design Team ------ --------------------------------------------------------------- 14:00 Design Team Report 20m M. Bagnulo, B. Claise Matt: Internal metrics include protocol statistics MIBs Andrew: Also router MIBs [Audio offset: 1:16:30] [At end of presentation, general discussion about what internal metrics are, relative to passive metrics] Matt: You expose state variables that TCP has that you can?t observe on the wire. TCP for correct operation, has to measure the network. Just gathering that information is very powerful. Its not passive, because you can?t make the same observation on the wire and its not active because you?re not sending any packets that wouldn?t have been sent if the user didn?t click the button. Brian: This here is still some discussion of where that internal stuff goes. It looks like a good idea to define them within the passive document and see if that works. Matt: Internal is different from passive because you didn't send anything the user didn't ask for, and yet you're doing a measurement internal to the protocol that you can't do passively. Brian: Although you can estimate those. Matt: Both have the property that you have to retroactively determine what the test traffic was. IV. Individual Drafts ------ --------------------------------------------------------------- 14:20 draft-chen-ippm-coloring-based-ipfpm-framework-01 10m M. Chen [Audio offset: 1:27:00] Mike Ackerman: The implementation in IPv6, has it been approved? Tentatively we have prototypes for IPv4 and being determined for IPv6. Mike: Delay measurements require time sync, has that been a problem? M.: It collects the timestamp. [audio offset: 1:29:00] 14:30 draft-zheng-ippm-passive-gap-analysis-00 10m V. Zheng [audio offset: 1:39:00] N. Elkins: In future work, are you intending to do work, or are you soliciting volunteers? V: both N. Elkins: We can definitely help. Is the management controller included, or is that an LMAP quesion? Or am I making no sense? Phil Eardly: To my mind, LMAP agents can do passive measurements. Matt: But that's out of scope here (not in LMAP). At the time 2330 was written, we didn't know about the TCP performance model... a lot of things have changed. Passive measurement wasn't included because we were afraid of scope creep. What we need is an algebra for combining measurements. Passive measurements were left out because we were scared of the complexity and privacy implications. V: Are you still scared? Matt: I don't know. Al: One of the solutions is to encrypt everything, and that's going to have a profound effect on our ability to do anything more than with the patterns of encrypted packets in passive measurement. Brian: Lets do coffee later, we don't have time. Al: I'd rather have a beer... Mike Ackerman: Thanks for bringing this to the forefront and I volunteer to read the draft. Brian: Registry design was easier when we didn't put active and passive into the same box (and split it in three instead), and so we should probably do the same for 2330 Matt: There are other areas of 2330 which need work. Brian: N-dimensional approach to 2330 would be more productive, rather than doing a 2330bis. 14:40 draft-hedin-ippm-type-p-monitor-02 10m G. Mirsky [audio offset: 1:45:00] Brian: We're not considering new working group documents at this time. Once we reduce the number of items in our workload, we will consider new documents. But I recommend continued development and discussion on the list. Dave Taht: +1 on working group adoption 14:50 draft-elkins-ippm-pdm-metrics-02 10m N. Elkins [audio offset 1:54:45] Al: Packets can sit on TCP's send buffer for a while... at what point do you put an accurate timestamp on it? Before entering the queue or after? N: You can do either or both. The application can generate stamps. I have prototype running code. You can have two different choices, or both. Matt: So you have significant amounts of non-TCP traffic? N: We wanted to cover everything. Matt: How do you use the IP-ID for diagnostics? N: It makes packet traces easier to read, is the short answer. If/As time permits: draft-deng-ippm-wireless-00 5m L. Deng