Minutes for IPPM @ IETF 85. ============================ Al Morton acted as Chair for this meeting. 58 people attended in person. Minutes derived from notes taken by Brian Trammell and Nevil Brownlee. There were several proposals that likely support the new LMAP initiative discussed everywhere/all-the-time for 3 days. ------------------------------------------------------------------------- 1. Advanced Stream and Sampling Framework. (10') Al Morton, Joachim Fabini draft-morton-ippm-2330-update-00.txt The New Proposal to update RFC2330 framework included results from reactive networks and was garnered interest on the list and at the meeting - need more review and updates before adoption. There was a question whether we should do 2330bis or an update, Matt thought that there was enough needing change for a full text bis. Discussion: Kostas: You're doing this on a cellular network, right? 3G. Would be interesting to see LTE. We have some work to discuss later on. Two have read draft. LMAP is driver. Kostas: you need a wireless guy. I can help. Matt Z on Jabber: goal of 2330 bis: an applicability statement? new metrics? Al: we'd call it an update. ------------------------------------------------------------------------- 2. Passive and Hybrid Measurements using IPPM Metrics (15') Brian Trammell draft-trammell-ippm-hybrid-ps-00.txt The Hybrid passive-active work has also seen interest on the list and can benefit from development of different "hybrid" types, calibration, and assessment of unauthorized traffic - especially during capacity measurement. - need more review and updates before adoption. Discussion: - Data from multiple observation points along a path i.e., what does 'hybrid' really mean? IPFIX for IPPM, at least for event-driven measurements Matt: need a way to detect unauthorised traffic ("adjunct measurement") Calibration also needed. Michael: Would this need extra hardware? passive gives a good background estimate. Mike Gugenheim: Impact of precessing 'at end?'? need to define accuracy required in practice. James Miller: How to derive baseline from several datasets with varying accuracies? Al: need applicability material for this ------------------------------------------------------------------------- 3. Model Based Internet Performance Metrics (15') Matt Mathis draft-mathis-ippm-model-based-metrics-00.txt The Model-based metrics draft has also seen interest and comments on the list. This proposal benefits from years of considering the BTC measurement problem. 4 or 5 had read - need more review and updates before adoption. Discussion: volunteers to read: Brian (Al: already support for this) ------------------------------------------------------------------------- 4. Curating Internet Measurement Data (5') Matt Mathis draft-mathis-ippm-data-curation-00.txt On Data Curation, Matt needs help with this, but there are many examples of big data collection efforts to draw on. Brian Tramell and James Martin volunteered to help. Discussion: Volunteers to read: Brian, Nevil James Miller: as we scale our program, different programs take up the model, how does metadate scale. look at github. Mike B.: Environmental + calibration data necessary Matt M.: adjacent logs by other orgs, data prov. Mike B.: key environmentals: special events. giant data center moves. elections. ------------------------------------------------------------------------- 5. Network Performance Measurements for IPsec (10') Yang Cui draft-bi-ippm-ipsec-00.txt The IPsec measurement proposal may have some merit, but needs further review. Discussion: Al: we had a study of introducing TLS in TWAMP, but no interest in WG. http://www.ietf.org/proceedings/73/slides/ippm-3/ippm-3.htm Let's test that again. Three have read the draft, one author. Yakov Stein: I agree with everything, but this could be solved vacuously via tunnels. Why do we need this? Cui: No tunnels in measurement environment. Kostas: opening another tunnel measures that tunnel. consider wireless environments, where another tunnel is not what you want to do. Sumita: Usually, TWAMP is over data service path, I assume that has ipsec. Cui: may be a reason tu use different keys (?) Sumita: why do we need twamp-only keys Kostas: We are actively looking for folks to work with us. Silence. Al: Ask on the list, not ready to ask for adoption. ------------------------------------------------------------------------- 6. Flow-based Performance Measurement (10') Yu Fang draft-sun-ippm-flowbased-pm-00.txt The Flow-based performance performance measurement draft did not generate any interest on list or at the meeting. No comments during discussion ------------------------------------------------------------------------- 7. Rate Measurement Problem statement draft (5') Al Morton draft-ietf-ippm-rate-problem-00.txt On the Rate Measurement Problem statement WG draft - it appears we need to add the TCP measurement requirement - everyone agrees that UDP cannot assess TCP Capacity in general. Also a suggestion to address the dynamic/reactive network behaviour as a feature or requirement. We do not want to encourage multiple TCP streams for testing or users. Al: separate TCP aspect, leave for model based metrics, or assume that model-based is subset of TCP testing... Yakov: we brought this same question up a couple of times -- missing a lot of TCP effects. we handled how to measure TCP loss rate, but tiny part of the problem. is being addressed. should be in the problem statement. There is other work being done right now (Yakov, David, Bob): what happens with constant bitrate sharing constriction with TCP? Matt M.: I suspect huge overlap. Morph two docs into one doc. Al: Problem statement separable. Matt's draft solves problem. Matt: This was the problem the WG was chartered to solve. Al: Perhaps LMAP could start out as IPPM subtopic. Henning: Quick remark: variable speed links are surprisingly common. Constant rate links exception as opposed to rule. In other countries, it's quite common to do different rates for UDP and TCP. Abilities to extrapolate: really big asterisk. Al: First draft we looked at allows us to expand metrics into consumer space Yakov: a lot of small tcp flows never reach equilibirium, long-term flow is now competing with all these little ones. more complicated problem. do we have to capture that Matt: history is littered with a lot of tech that fundamentally clashed with tcp. easy to construct networks that deliver udp at finite loss rate. some of them even got fielded. ability to deliver udp packet not nec. useful. tension about web content sharding across mult. servers. spdy works hard to undo that. imagine web working on single streams. ------------------------------------------------------------------------- 8. RFC 2679-bis (10') Al Morton One draft was approved in the Interim, RFC 2679 tests supporting advancement. We have new text for RFC 2679bis with agreed modifications - need WG review of other small updates adding references to related work, then adopt/WGLC. Discussion: One reader. That's not enough. Matt: Update 2330? Imagining inserting more text, or 2330 bis? Al: add text, header pointer in 2330. Does 2330 need a bis? Matt: a lot of stuff we didn't know about, would do differently. one of the problems with BTC: doesn't tell an ISP what they need to do to meet it. well defined has two aspects: looking up from customer, looking down from NOC, some metrics in 2330 would help to look to see if we might want to change things. Al: what we need now is rechartering. Yakov: was happy about this meeting. last meeting i thought we were closing. Al: It feels like 1997 again,... Matt: re scope: at the Danvers BOF, it was IP provider metrics, charter was narrowed. Matt: good to stay out of NOC metrics, but C&C could be in scope. Matt has original notes and BoF charter.