Minutes for the MIPv6 Signaling and Handoff Optimization BOF (mipshop) Minutes taken by: Eva Gustaffson, Spencer Dawkins, Koojana Kuladinithi with some edits by Gabriel Montenegro Wednesday, July 16 at 0900-1130 ================================ CHAIRS: Basavaraj Patil <Basavaraj.Patil@nokia.com> Gabriel Montenegro <gab@sun.com> Phil Roberts <proberts@megisto.com> 0. Executive Summary: Initially, the discussion centered around the charter items. Folks suggested other potential items like smoothing techniques (e.g., bicasting) and network-controlled handoffs. However, initial focus will be on finalizing HMIPv6 and FMIPv6 specs for experimental status. Nevertheless, it is important to keep an eye on other groups (e.g. seamoby, the irtf mobility work). So further work remains possible (but may require rechartering). Greg Daley gave us an overview of the DNA (detecting network attachment) BoF scheduled for Friday, with an emphasis on its relationship with MIPSHOP. This refers to items like movement detection and fast address configuration on the new link. Further details need to be worked out, but this was a good heads up. Rajeev Koodli presented the lastest changes to the FMIPv6 draft. In particular, this concerns the new mechanism of a fast NA (FNA) encapsulating an FBU. Some suggestions that this needs airing within the IPv6 WG, but it was also mentioned that this type of mechanism is ok for an experimental draft. Hesham reviewed the latest changes to HMIPv6, in particular, the on-link care-of address test so the MAP can (optionally) perform a return routability test of the MN's LCoA. He also went through the LMM requirements draft to verify that HMIPv6 satisfied it. Some requirements are unclear (e.g., "not introducing a single point of failure" -- which the MAP, like a HA, is). This may require revising LMM Requirements. We then had 3 implementation reports. Rajeev Koodli talked about an experimental FMIPv6 testbed in which they are measuring packet loss rates of voice traffic during fast handoffs, comparing with amount of buffering. Jim Kempf reported on some simulations of up to 100k users, and Greg Daley reported on the Monash HMIPv6 implementation. Next steps for the BoF/WG is to send the LMM Requirements draft to IESG for informational, and target Oct for WG last calls on both HMIPv6 and FMIPv6. Status of WG I-Ds (MIPSHOP related) draft-ietf-mobileip-hmipv6-07.txt draft-ietf-mobileip-fast-mipv6-06.txt draft-ietf-mobileip-lmm-requirements-03.txt All are work in progress ---------------- DETAILED MINUTES ---------------- 1. Charter discussion 15 Chairs MIPSHOP BoF charter discussion HMIPv6, FMIPv6; use these as basis for experimental protocols, also refine LLM requirements, follow and work with other relevant efforts ex IRTF group on mobility Hesham Soliman: security for FMIP? Gabriel Montenegro: yes, it was there Hesham Soliman: got impression we would go for experimental without look at security Gabriel Montenegro: yes, we go for experimental but well look at security too Alper Yegin: separate draft? Gabriel Montenegro: possibly Karim El-Malki: Will smoothing techniques be part of the charter? Gab: open for discussion here, but initial plan is existing drafts. James: we have three drafts. We should finish these. If we have more drafts, bring up the issue in the WG, dont decide this now. Gwendowoo?: would like to see network-controlled handoff included. Not just layer-two attachement. Charlie: Need to tie HMIP, and maybe more, to SEAMOBY. Raj: lots of other drafts. Not discouraging them, focused on these three for now. Next steps: Sep 03: LMM requirements to IESG Oct 03: WG last call on FMIPv6 Oct 03: WG last call on HMIP Nov 03: discuss last call comments at IETF 58 2. DNA interaction 10 min Greg Daley no draft Running a BoF on Friday this week on Detecting Network Attachment. MIPv6 optimizations: movement prediction (FMIPv6), movement detection (MDOpt, FastRA, FRD...), address configuration (OptiDAD, aDAD...), location signaling (HMIPv6) Four applicable areas FMIPv6 and HMIPv6 are in MIPSHOP, other two arent why not? Rajeev: Taxonomy is good but not perfect... Overall solution to DAD opt may be good solution to look at for DNA Charlie: DAD isnt really required, we just dont know how to avoid it for lots of links. Movement prediction isnt that close to fast handoff. Greg: youre probably right, I'm over-simplifying here. If you dont have any information about the network youre going to, movement detection does help. DNA BoF is about either quickly doing configuration or figuring out that we dont have to do configuration. Rajeev: why are you restricting md to handovers? We already have a handover solution in MIPSHOP. Greg: this presentation is about handovers, the BoF wont be. We want to include unmodified routers in DNA, too. Could receive router advertisement quickly, or base movement detection on simple layer two triggers. Jitter: its not clear how a mobile node would make sure its router is still there. Greg: no interest expressed in MIPSHOP want to check this in larger context. Is Duplicate Address Detection optimization worthwhile? Where should it live? [note: output of DNA bof seems to indicate that DAD optimization will be homed in an upcoming DNA WG] 3. FMIPv6 update 15 Rajeev Koodli draft-ietf-mobileip-fast-mipv6-06.txt Have had feedback on three-party handover (do two-party first!) and L2-optimized handover. Each of these may make sense as separate draft. Two scenarios on why MN might need to do Fast Binding Update. FBack_Count is zero in both scenarios. Send FBU and FNA (requesting Reachable transition at new access router). Hesham: Reachable is supposed to mean you've tested bidirectional reachability. Rajeev Koodli: good point, you dont have bi-directional reachability confirmation at this point, ... we could set it to stale and go from there, maybe? Or router may already know, so it COULD be reachable. Hesham Soliman: its not a downside that router will do this in parallel, we could optimize neighbor discovery but we should discuss this in MIP6 WG. IPv6 spec says specifically not to set reachable based on layer-two information. Rajeev Koodli: yes, ... James Kempf 2461 predates widespread availability of wireless. DNA needs to fix it. No problem with experimental draft. ...Cont presentation: NAR could verify NCoA is conflict-free while creating new forwarding entry. ... FBU transmission from new link FNA should be changed to Destination Option. Hesham Soliman: thought you wanted to have double IP packet encapsulated...? Rajeev koodli: base IPv6 packets, .... two IP packets Have also received feedback on HI and HAck, necessary to avoid ND, but requires buffering of incoming packets 4. hmip [and fmip?] vs lmmrequirements 10 min Hesham Soliman draft-ietf-mobileip-hmipv6-07.txt draft-ietf-mobileip-lmm-requirements-03.txt No clear reason to distinguish between MAP and HA, but optional LCoA test (OCoT) was added just in case MAP turns out to have stronger security requirements. All messages authenticated with ESP/AH. Lots of LMM requirements, and some are unclear. Only requirement not satisfied by LMM is avoid introducing single point of failure in the network. MAP is single point of failure, just like HA. James Kempf: change requirements to avoid introducing additional points of failure? Hesham Soliman: yes... 5. Implementation reports: 20 min Fast handovers implementation experience (Rajeev Koodli) Experimental testbed, linux MN, 802.11b, free-BSD derived router Experimental Fast Handoff testbed using 20-ms 24-byte talk spurts. Have to buffer or expect to lose about 15 packets during handoff. No packets lost unless FPU is lost. With 500-ms duration, lose at least five packets about a third of the time. Heshom: is this a simulation? Rajeev: No, its a testbed Hesham: Do you know how often FPU is lost Rajeev: It's traffic dependent For existing sessions, theres no ramp-up. You have to continue stream without heads-up signaling. Pete McCann: How much delay is introduced by this buffer? Sufficient buffering is indistinguishable for voice applications. We need to fix handover to use these networks for voice. Gopal: Im shocked by your loss. Sometimes youre losing all the packets? Rajeev: router buffering removes constant inter-packet arrival. Alister: voice has a big tradeoff between buffering and delay. This is the same tradeoff we were making on packet radio in the 1980s. Rajeev: definitely traffic-dependent. ---------------------------------------------------------- Scalability of FMIPv6 and HMIPv6 (James Kempf) Piecewise simulation procedure, 100K users simulated Conclusion: FMIP should be simplified to reduce amount of over the air signaling associated with IP handover Hesham Soliman: with HMIP you assume no anticipation? James Kempf: results combination of anticipated and not anticipated ---------------------------------------------------------- Monash HMIPv6 implementation report (Greg Daley) (No questions, ran out of |