Minutes of the IP Performance Metrics WG (ippm) Monday, July 15 at 13:00 - 15:00 ==================================== CHAIRS: Merike Kaeo , Matt Zekauskas Al Morton took notes, which were edited into these minutes by the chairs. AGENDA: 1. Agenda Bashing 2. Working Group Milestone Status 3. IPPM Reporting MIB and Metrics Registry Discussion 4. Packet Reordering Discussion 5. One-way Delay Protocol Requirements Discussion 6. One-way Metrics Applicability Statement IETF home page: http://www.ietf.org/html.charters/ippm-charter.html IPPM home page: http://www.advanced.org/IPPM/ 3. IPPM Reporting MIB and Metrics Registry Discussion -- Emile Stephan http://www.ietf.org/internet-drafts/draft-ietf-ippm-reporting-mib-00.txt http://www.ietf.org/internet-drafts/draft-ietf-ippm-metrics-registry-00.txt Emile Stephan presented on the metrics registry and reporting MIB. First, the metrics registry document was discussed. The document defines a registry format, associating OIDs with IPPM metrics. RMON would like the registry so there are unique, common, ways of referencing the IPPM metrics. One issue yet to be resolved is how to define new metrics. This draft has a draft tree, where draft metrics can be placed to avoid ad-hoc assignment. The names would have to be changed when the metrics move into the "rfc" (standardized) branch. Emile is also placing metrics from IPPM drafts in the IESG last-call process into the standardized portion of the registry. Another issue raised is the metric OID length. He is in discussion with SNMP experts on this issue - one home node for the IPPM Metrics Registry could allow the metric object identifier to fit in exactly 8 bytes. There were no comments from the group on these issues. Next, Emile reported on the status of, and outstanding issues with, the reporting MIB. The first major issue is with timestamp resolution. His current proposal uses timestamps with 250 picosecond resolution. Is that enough for the next few years? Henk Uijterwaal made the comment that the proposed timestamp looks like it has the year 2100 problem - only two digits are displayed. He advocated using timestamps from NTP; then have multiple resolutions and all the tools exist for conversions. Jon Bennet supported this idea. Emile made a comment that you need something that could be used as an index, and you must have GMT not local time. You also need something human readable. Henk noted that NTP timestamps are GMT: tools exist for human-readable forms (also to convert from GMT to some local reference) and that as a 32 bit number it should be a suitable index. Emile will look into the issue, and there will be further discussion on the mailing list. Next, Emile noted that he added a field for synchronization (is clock synchronized or not?). A question from the floor asked about accuracy - Emile responded that the accuracy was in the MIB. Measurement management issues were raised. Is it useful to have history when one measure was more precise than other? Some seemed to agree that this was useful. Emile proposed SNMP over TCP as one way to help secure connections, so that you could control the measurements; he also made the point that it is important to have measurement packet interoperability. Matt Z. made statement that giving Emile feedback on control is fine, but it's not what we're doing in the working group. Al Morton noted that packet format is exactly what OWAMP is addressing. 4. Discussion on Packet Reordering -- Al Morton http://www.ietf.org/internet-drafts/draft-ietf-ippm-reordering-00.txt Al Morton presented status of current draft: it is a combination of the three drafts presented at the previous IETF. MLAS needs complete sequence and doesn't work if packets are lost, which is problematic, so it is not considered in this draft. (MLAS did have good property of identifying how many steps needed to restore order; it is just not clear how to apply it). The authors decided on two goals: first, determine whether packet order is maintained (and which packets are reordered). Second, quantify the extent of change. Al outlined the changes yet to be made, the principle one being the use of consistent notation. He also went over an extensive example showing the metrics in use (one comment was to make sure the examples in the draft show the application of all the metrics). Stanislav noted that the vector of arrival order in the example shown on the slides was not represented correctly - the notation should be fixed. Notation in the example section of the document should be checked too. Merike asked how many people read draft; about 15 people raised hands. There were no other substantive comments from the group. Merike asked people to take and further comments to mailing list! 5. Discussion on One-Way Active Measurements Protocol Requirements -- Stanislav Shalunov http://www.ietf.org/internet-drafts/draft-ietf-ippm-owdp-reqs-02.txt see also http://www.ietf.org/internet-drafts/draft-ietf-ippm-owdp-04.txt Stanislav Shalunov presented the modifications to the OWAMP requirements document. In general, there were just textual modifications and small clarifications. Stanislav noted that the authors thought it was ready for publication. Matt took a rough consensus poll: approximately 5 had read the draft, and of those, all thought it should be moved forward. As a check Matt asked how many were interested; in the room approximately 10 raised their hands (most of the rest appeared to be lurkers). Al asked if the timestamp resolution comments from December were taken into account. Stanislav said that the timestamp resolution itself does not belong in requirement document. After discussion, Al was satisfied, but will reread the draft. Matt said the chairs would call for WG last call after the IETF. Stanislav then presented the changes in latest protocol draft, as well as pending changes. He asked for comments on pending changes before they actually go in the next version of the draft. A person from the audience (Sharam) asked a general question: why not use ICMP time stamps with a reliable return channel? You can then do everything from just the source. Stanislav pointed out that in abstract there was no problem with ICMP packets (as long as you have synchronized clocks), but you can't use them for accurate performance measurement. ICMP is not reliable, not treated like other packets in all networks, it may be identifiable, and it needs processing time at the responder (which may affect accuracy). With ICMP you can also not guarantee that the forward path is the same as the reverse path. The questioner mentioned that perhaps there could be some kind of traffic-engineered reverse path to "ensure" reliability. 6. One-Way Metric Applicability Statement -- Henk Uijterwaal http://www.ietf.org/internet-drafts/draft-ietf-ippm-owmetric-as-00.txt Finally, Henk Uijterwaal led a discussion of open issues with the one-way metric applicability statement draft. He started by noting that the draft should have a negative version number - it attempts to lay out the issues, but doesn't provide lots of answers. The intent is to use it as a basis for discussion; input is especially requested from other operators. The first issue was sending rate. Stanislav noted that the rate really depends on what you want to measure; looking for delay, one value is appropriate, for reordering you would want the rate to match a target application, and for capacity you want a much higher rate. He didn't see any single answer. (There was a "3% of link capacity discussion, which will be reported in a subsequent paragraph.) John Bennett noted that the rate needed might vary by time of day as well. Henk agreeded, but noted that bandwidth measurements certainly could swamp a link, although it might be appropriate if done infrequently. The next issue that generated discussion was packet size. Al Morton and Stanislav Shalunov were among the contributors. Again, the size depends on the purpose. IPDV requires one packet size, but that's only for a particular variation; there's no reason you can't calculate IPDV on different sized packet streams. Another issue was test duration. For delays, can do things continuously; for bandwidth you need to be careful and the test should be short-term. Al Morton noted that how frequently you report intermediate results determines how quickly you can act on it. He has a system that does this every 15 minutes, and that seems to be a good compromise. There might be some data loss in-between cycles, but it is a way to get fairly good coverage with a compromise. Another issue is data volume. 3% of link capacity was a figure from questioning providers. Note that it is a MAXIMUM over a LONG TERM. We are looking for providers that might violently disagree. You might also need to specify a distribution instead of just a blanket volume. The final issue that generated discussion for this document was security. Henk noted that there are potential DDoS problems: any single stream might be acceptable, but the total measurements might swamp a single receiver. (For situations when there are many senders to one receiver.) Stanislav mentioned that any kind of "amplifier" is not acceptable. For OWAMP, we're looking at authentication, and if a measurement point doesn't know you, then it will only send traffic back to you and not accept traffic from you. Finally, a group member asked about packet loss and errors. In the wireless area, very interested in loss due to errors instead of loss due to congestion. Was there a differentiation with IPPM? If you are looking at the network, you often don't have any observability into why a packet was lost. This question might be better addressed in tewg; Stanislav noted that the solution is sub-ip, so perhaps the measurement should be in sub-ip.