IRTF Open Meeting Quebec City, Canada Wednesday, July 27, 2011, 9:00 - 11:30 Room 202 -------------- State of the IRTF RG Chairs & Lars Eggert Lars presented the IRTF status update - slides: http://www.ietf.org/proceedings/81/slides/IRTF-6.pdf Feedback was solicited on how to arrange IRTF presentation to the Technical Plenary vs. IRTF Open Meeting agenda. Research Group reviews - slides as before: ASRG - little activity, planning to meet in conjunction with MAAWG to rectify this. CFRG - helping IETF move towards crypto policy appropriate for modern age - meeting with IAB this week - co-chair stepped down - looking for a new co-chair - looking to stimulate more activity on validation and assurance DTNRG - started out as interplanetary research group - DTNRG now a little closer to earth ;) - protocol has been used in space - Stephen Farrel has cool project using this in swedish military applications HIPRG - wrapping up their main charter item, will be discussing the future of the group at their meeting tomorrow ICCRG - incorporating transport layer topics from end-to-end RG that closed MobOpts - forum for discussing alternative mobility mechanisms, e.g. multipath TCP - interested in trying to find a community around these questions to make progress, otherwise will shut down - will be touring academic conferences soliciting contributions NMRG - has new chair - ran a poll on the mailing list to identify future topics for research P2PRG - rechartered about 2 years ago - since then have published 2 RFCs - p2p security, and alto techniques survey - traffic localization is current focus RRG - have been incredibly busy for last few years - work finished, now in recovery period - looking for new topics - if there is sufficient energy to continue will recharter, otherwise close SAMRG (not presented) TMRG - moving this work back into ICCRG VNRG - Martin Stiemerling presenting - RG founded 18 months ago - working on diverse field of virtual networks - seeking to define their terms for virtual networks - diverse set of people - hard to get them to do work between meetings - have good discussions during meetings - trying to get people back on board, stimulate discussions between meetings - still a lot of talking past each other - will be meeting on Friday morning this week --------------- Applied Networking Research Prize (ANRP) Awards - see http://www.irtf.org/anrp for details *** Mattia Rossi *** for research into reducing BGP traffic: A Technique for Reducing BGP Update Announcements through Path Exploration Damping. Geoff Huston, Mattia Rossi and Grenville Armitage. IEEE Journal on Selected Areas in Communications (JSAC), Vol. 28, No. 8, pp. 1271-1286, October 2010. Abstract: Path Exploration Damping is intended as a replacement for MRAI and RFD, the two main mechanisms used in BGP that aim at a reduction of BGP update churn. This talk explains the motivation behind PED, outlines the algorithm and provides insight into experimental results. Questions: Mat Ford - Why set different PEDI per peer? or per prefix? What indicators can an operator use to tune these parameters? Mattia - this is the next focus of our research, we're studying per prefix behavior to understand better what could be useful in this regard and to develop heuristics. ? - Any difference between iBGP and eBGP? Mattia - Yes, we're only looking at eBGP. There is no timer for iBGP. People not always happy about this - more updates in iBGP then eBGP. Maybe in future we will recommend a lower default if other defaults are reduced, could become automatic. Georgios Karagiannis - interested in experiments and statistical accuracy. mentioned that relative times are quicker, but have you used multiple rounds? Mattia - this is exactly why we used the accelerator - to replay the data set multiple times - the results I showed you are the averages. GK - What about confidence intervals? Mattia - I think these are documented in the paper. ----------- *** Beichuan Zhang *** for research into "green" traffic engineering: GreenTE: Power-Aware Traffic Engineering. Mingui Zhang, Cheng Yi, Bin Liu and Beichuan Zhang. Proc. IEEE International Conference on Network Protocols (ICNP), pp. 21-30, October 2010. Abstract: Given the high path redundancy and low link utilization in today's large networks, one may route traffic using fewer number of links so that other links can go sleep to conserve energy. In this work we design a traffic engineering mechanism that maximizes the number of links to sleep under certain performance constraints such as link utilization and packet delay. Using network topologies and traffic data from several wide-area networks, our evaluation demonstrates that GreenTE is both feasible and beneficial. Questions: Dirk Kutscher - aware of approaches for wireless networks - slightly different problem - there the observation is that you save the most power not by turning base station, wireless i/f off, but turning whole site off, e.g. air conditioning etc. = real problem is when to switch on again. Beichuan - We didn't consider that in this work - these are large routers, we're assuming they stay on - with better support for whole platform sleep modes we could explore this Dave Crocker - this is interesting work - you're taking a whole system view, focussing on line cards - what are total savings relative to total consumption? Beichuan - line cards consumption is 40-50% of total power consumption, also have data that says 40-70% - so significant amount of total power consumption could be saved - we only look at line cards, but i imagine there may be indirect benefits, e.g fans could be turned off once number of line-cards drops below some threshold JinHyeock Choi - server has perfect knowledge of traffic matrix? Beichuan - traffic matrix is collected from ingress routers - then we estimate traffic matrix for next interval JinHyeock Choi - so you predict for the future? Beichuan - yes JinHyeock Choi - what if there is a discrepancy between reality and prediction? Beichuan - power savings may not be optimal, but this doesn't violate constraints JinHyeock Choi - slide 22 - power saving potential drops rapidly, why? Beichuan - lack of flexibility in topology means there are inflection points where lots of links have to be enabled simultaneously Georgios Karagiannis - transition from sleep to active mode requires a power surge - did you include this? BZ - we didn't do this comparison - when we turn off a card we let it be off for a long period of time - not opportunistic off/on - can amortize this power over time by making power off intervals long, but the point is valid. Lars Eggert - it happened that both prize winning papers were in the routing area, that is coincidence, there are no such restrictions an ANRP nominations Mat Ford - deadline for next call is aug 28th Lars - email will go out soon with details of how to nominate etc., please nominate! ---------------- Presentation & discussion of proposals for new IRTF research groups Network Complexity Michael Behringer http://networkcomplexity.org/ This would be extremely valuable for security analysis Bob Briscoe - there are limits to how complete an analysis can be arrived at Michael - yes ---------------- Information-Centric Networking Dirk Kutscher & Borje Ohlman http://trac.tools.ietf.org/group/irtf/trac/wiki/icnrg - no questions ----------------- Learning-Capable Communication Networks Dimitri Papadimitriou https://sites.google.com/site/lccngroup/ Marcelo Bagnulo - we have routing protocols - they self-heal already - so can you provide a more detailed example to motivate this? Dmitri - please come this afternoon to our meeting, where we will show that instead of adaptive routing we have self-adaptive routing Lars - why not do this in existing NMRG? Dmitri - the role and scope of self-management as indicated on slide 5 - bounds are enforced on a global configuration, based on policies determined by operator's high and low level objectives. Lars - I'm wondering if the broader nm community subscribes to your compartmentalization of the activities here. Dmitiri - had this discussion in the last meeting - what is the value of enabling this system with a learning component is first question - there are advantages in allowing systems … Lars - I understand that you don't think it fits in NRMG - does NRMG agree with that? ? - not really - for several years we have had workshop on flow-based techniques in NMRG - typically these kind of things are already addressed in the group Lars - can we have this discussion in the NRMG is the question? let's continue this at the meeting this afternoon. ----------- Energy Efficient Communication Rolf Winter & Juergen Quittek No further questions. Meeting adjourned. -----------