IRTF Open Meeting Taipei, Taiwan TUESDAY, November 15, 2011 1300-1500 Afternoon Session I Room 201 ABC State of the IRTF (RG Chairs & Lars Eggert) Lars and various RG chairs presented status updates IPR and the IRTF Lars presented clarification that IETF "Note Well" does apply to IRTF participation. Explicit document explaining this is being prepared. Ken Carlberg: How is NCRG proceeding? Lars Eggert: Meeting is planned for Paris. Work has been done already. There is a wiki. Applied Networking Prize (ANRP) Award Lars presented background on ANRP Award. Nomination form is at Nominees need to have published or accepted academic paper. Both self-nomination and third-party nomination is encouraged. *** Michio Honda *** for his research into determining the future extensibility of TCP: Michio Honda, Yoshifumi Nishida, Costin Raiciu, Adam Greenhalgh, Mark Handley and Hideyuki Tokuda. Is it Still Possible to Extend TCP? Proc. ACM Internet Measurement Conference (IMC), November 2011, Berlin, Germany. Mat Ford: Do you intend to continue taking measurements to see how these results change over time? Michio Honda: Peer Azmat Shah: Is head-of-line-blocking a problem? Michio Honda: No Matthew Mathis: What software did you use to send raw packets from Python? Michio Honda: Raw IP Sockets on Linux, libpcap on BSD and OS X Hannes Tschofenig: It may be difficult to do the same tests repeatably over time. Also, you meshed two things together in one document: 1. New TCP Encryption option, 2. Measurements. Would like data on how IP options fare in today's Internet. Can't we have a more systematic way of taking these measurements? Michio Honda: The Internet evolves unexpectedly Lars Eggert: This is not the only work measurement work going on. This was presented at IMC -- Internet Measurement Conference Matthew Mathis: These tools require root permission Google is doing an IPv6 experiment in JavaScript via YouTube views, which does not require root permission. Stuart Cheshire: JavaScript experiments in the browser are useful, but some experiments can't be done that way Spencer Dawkins: I'm really pleased with the award being given for this work. There used to be a Working Group working on stuff like this. Maybe we could resurrect that. Dirk Kutscher: Look into different architectures like home networks and 3G. Stuart Cheshire: Different things are possible on different networks -- things like IPv6 link-local can be used on a home network but not the public Internet. Knowing what's viable on the public Internet is still very valuable. HK Chu: Should have an RFC or other document on what is the right thing to do for new vendors. E.g. for TSO or Large Segmentation Offload. Matthew Mathis: The IETF needs a doc stating what is allowed and what is not. Otherwise we have no way to tell vendors they are wrong. Piers O'Hanlon: Can you use something like nmap to identify what middlebox did the damage? Michio Honda: No Hannes Tschofenig: Some work has been done on identifying middleboxes. Will you release the data you collected? Michio Honda: Yes, the URL for the raw data is in the paper, with anonymized IP addresses *** Nasif Ekiz *** for his analysis of misbehaving TCP receivers: Nasif Ekiz, Abuthahir Habeeb Rahman and Paul D. Amer. Misbehaviors in TCP SACK Generation. ACM SIGCOMM Computer Communication Review, Volume 41, Issue 2, April 2011. Mat Ford: Are there any conclusions you draw about how we might specify future IETF standards to reduce such mistakes? Nasif Ekiz: Implementers should test better Yoav Nir: But TCP does actually work on Windows. How much performance degradation did all these problems cause? Nasif Ekiz: In some cases, half Brian Trammell: Were you testing just single byte drops? Nasif Ekiz: No. It's just written that way for clarity in the paper. Spencer Dawkins: Thank you for doing this work. You talked about the difference between unexpected behaviors and misbehaviors. At the IETF we've seen spectacular misbehaviors. I hope you are presenting this work elsewhere too. Nasif Ekiz: Most of these things were "SHOULD" statements in the specs, but they still look like bugs. In some cases we could look at source code to see the bugs. Matthew Mathis: Every single SHOULD is there because there's also a reason not to do it. Reneging language was there specifically to accommodate receiver bugs. Dave Oran: Where there was state leakage between TCP connections the example had the same ISN for both connections - which should never/ rarely happen if you implement ISN randomization correctly. Ning So, Cross Stratum Optimization Hannes Tschofenig: Are you planning to look into the specific needs of particular applications? Ning So: We don't want to attach the study to any specific application. Hannes Tschofenig: Different applications have very different needs. Susan Hares: Would you like to get cooperation from people like game developers? Ning So: Yes, we would love to have collaboration.