Minutes of the IP Performance Metrics WG (ippm) Tuesday, 12 December 2000, 17:00-18:00 ====================================== The meeting was chaired by the working group chairs, Matt Zekauskas and Merike Kaeo. Thanks to Paul Love who took notes during the meeting. AGENDA 1. Introduction of new co-chair & agenda bashing 2. Discuss the new charter and milestones 3. Update on Bulk Transfer Capacity 4. Update on Network Performance Measurement for Periodic Streams 5. One-way delay protocol document discussion 6. Discussion of any issues with IPDV or loss patterns IETF home page: http://www.ietf.org/html.charters/ippm-charter.html IPPM home page: http://www.advanced.org/IPPM/ 1. Introduction of new co-chair & agenda bashing -- Matt Zekauskas, Advanced Network & Services The meeting was called to order by Matt Zekauskas who introduced the new co-chair and reviewed the meeting agenda. 2. Discussion of new charter and milestones -- Matt Zekauskas A discussion on the new proposed charter and new milestones was led by Matt who noted that what's on the web page is 3 years old and stale. Matt read/editorialized the new text which was published to list. The following points were made: - Draft on protocols will be a priority in upcoming weeks - The Framework, RFC 2330, was reviewed by the Chairs and looks to still be OK. Therefore it is pulled from milestone list and people were asked to post to the list if they thought otherwise. - Need a BCP on one-way delay & loss. - Need statistician(s) to help with comparisons & validity of metrics. Henk Uijterwaal commented that Les Cottrell and a SLAC colleague have done some work on this (see the Proceedings of the The First Passive and Active Measurement Workshop (PAM 2000), Hamilton, NZ, April 2000). - Longer-term milestones, where nobody has signed up, will be deleted if no one steps forward. These include: Mar 01 Develop draft for comparison and validity of metrics Jun 01 Develop draft for passive application of metrics Nov 01 Develop draft for BCP to apply existing metrics to diffserv Nov 01 Develop draft for BCP to apply existing metrics to multicast Comments from floor on milestones: Will Leland noted that IPPM also needs to do work on MIB(s) for metrics. Bob Cole stated there is work being done on a MIB to instrument inflow. RMONMIB WG is doing work on additional, related MIBs so perhaps no additional work is needed. Matt Zekauskas commented that IPPM would like to work closely with that effort, and is also trying to work closely with the Traffic Engineering WG. [After the meeting two people approached the chairs saying they were statisticians and were interested in pursuing the comparison draft. The chairs will follow this up with them (one from Telcordia, one from CAIDA).] 3. Update on Bulk Transfer Capacity -- Mark Allman, NASA Glenn Mark used one slide to say that the authors think the framework is done and the question was posed to the group as to what they thought. It was asked if this was a draft or a framework? The answer was that it's a framework to define BTC that tries to clear up which things need to be nailed down by implementation. In January, one particular implementation will be published as a draft (see milestone list). There was agreement in those present who have read that it should go out. 4. Update on Network Performance Measurement for Periodic Streams -- Vilho Raisanen, Nokia This draft is now in 3rd revision with changes from 2nd being relatively minor. There was a comment that it does not address statistics - this it is just a building block. It can also produce results on packet loss. There were comments made with respect to implementations, principally that having an implementation would be desirable before advancing the draft. Vilho reported on a draft of an implementation report. It is draft-raisanen-ippm-npmps-results-00.txt; a PDF version is available. This implementation draft describes application of the metric. Also, packet integrity checks will be added into the draft to account for packet errors. There was a comment related to the figures from the implementation draft. Vilho described that the delays shown in the figures were not one-way, but estimated by taking half of round-trip delay. The two delay distributions in the second figure were taken from very different networks: one a private network and the other a public Internet, hence differences shown. Matt Zekauskas asked if including what a Poisson stream would have show for same networks would be useful? Merike asked if comparison data between Poisson-type tests and the stream-type tests could be included. 5. One-way delay protocol document discussion -- Stanislav Shalunov, Internet2/UCAID Stanislav reviewed the protocol's goals and stated that this protocol was designed to be useful in a wide range of scenarios. He addressed some comments from the list including making padding pseudo-random and arbitrary. He also clarified that the Test protocol can be used without the Control protocol. In the ensuing discussion, Matt Mathis commented that the word 'stealth' should not be used because of some potentially bad connotations due to the way the word is used elsewhere. Also, he added that for the test packet you could create an opaque cookie and hide the test packet in a more generalized TCP packet so there was no need to recreate packet structure. All you have to do is specify length and offset for the test packet. This would also make the protocol flexible for new uses. Ruediger Geib noted that a version number should be included for backwards compatibility, especially for the control protocol. He then asked how the protocol handles responses that are unexpected and suggested that the draft be more specific under which conditions packets are discarded. The built-in resiliency needs to be specified in the draft for implementers to all select the same mechanism. Ruediger also initiated debate on how start and stop messages were used. It was decided to take the issue onto the mailing list. Bob Cole commented that cell delay variation should be added. Matt Zekauskas would like any results (of cell delay variation studies, in particular on using periodic streams) published on list. Henk suggested extending test packets to allow more information than just time stamps. Henk will supply samples. Also, should the two protocols (Test and Control) be in one or two separate documents? Bob Cole thought that if in one document, then it would be easier to talk about how they were used in various situations. Stanislav said he would summarize his interpretation of what was needed to the list, and that others should comment. 7. Discussion of any issues with IPDV or loss patterns Phil Chimento, the IPDV author, thought IPDV was nearly done, but recently there has been lots of traffic on the list. He went over the list of comments made to the list and promised to have a new draft by 1 Feb 2001. Al Morton commented that specification of delay variation is simplified if packet size doesn't change. Phil thought this was added to one of the earlier versions but if not, he will add to the new draft.