March 14, 2013 9:00 - 11:30 Chairs: Mark Townsley & Ray Bellis Minutes: Thomas Clausen Jabber: Ole Troan & Jason Weil --------------------------------------------------- Introduction by chairs, reminder of Note Well, etc. Thanks to Ralph Droms for his tenure as INT AD, and welcome to Ted Lemon as new INT AD. Round of applause for both outgoing and incoming ADs. Mark Townsley reminds everybody that next week, the IEEE meets in the Orlando location, and as it is in the charter of HOMENT to interface with other SDOs, Paul Unbehegen is invited to come talk about IEEE 802.1aq Presents the agenda for bashing - no comments to the agenda. 9:08AM-9:34AM - Tim Chown - Homenet Architecture I-D update Gives update on the main changes from -06 to -07 Presented outcome of WGLC that ended March 4: - General support - A few suggested the document wasn't ready, with lists of things to consider; the authors have started folding those in and expect to produce a -08 - Also, the document is long, and so may need an effort to make it leaner - Some comments suggested that the principles / requirements were hard to find. Tim reminded everybody that there was, in an earlier version, enumerated lists but it was at the time requested be removed. It can be inserted again, though. Delegated prefixes (slide 4) were discussed. Ted Lemon: Suggests that it is not going to be "Flash renumbering", it's going to happen every day - so you can allow connections to die like over an hour Tim Chown: Depends on what happens to the old addresses, if they are still useful Ted Lemon: In the cases I am aware of, the prefix becomes deprecated, not invalid Jan Žorž: All of the RIRs recommend that the allocation policy be based on the 56, if you divert from that, need more than that, then... Tim Chown: Think that /56 is enough, happy with RFC6166 ??: In LACNIC we do not recommend a /56 or whatever, have a pointer to RFC6177 Lorenzo Colitti: Did we try to converge to come up with a recommendation, or was it too hard, or we didn't try. Tim Chown: We'd want to see at least a /60, a /56 is great Lorenzo Colitti: Will this lead to negotiation? Lee Howard: If we want to converge on a recommendation for allocation size, that has to be a different document. If we have a /60, we cannot have multiple routers. With response to the deprecation time, I would love to see an ISP that requires renumbering would update DHCP servers, and then keep the prefixes alive, both, for say 2 hours Peter Lothberg: Every time we've said "X is enough" we've been wrong. Given a /56 to the home, that's a lot of devices John Jason Brzozowski: Typically, we would lower the lease time well ahead of maintenance. As far as the text in the arch doc, do you feel that it is important to enumerate the conditions under which this occur, or? Should it be clear to the readers of the document why we renumber Tim Chown: The implications if you renumber every day, as opposed to every year, so... Peter Lothberg: Do not think that we should tie the architecture to DHCP Jan Žorž: IS the IETF the right place for this? Should the operators make the recommendations to the RIRs? Tim Chown: RFC6177 kinda says this, and we're threading in its footsteps, although the RIRs of course have the final word on what they do. Lorenzo Colitti: I've been trying to figure out how ULAs might work in a tree topology without a routing protocol. I do not think that we can make them work, at least we do not know how to make it work, in a sorta PD-tree. If it starts at the bottom of the tree, then it won't work due to the requirements that it be dropped at the "north end" of the tree. So we need some text saying that you must route ULAs properly. Tim Chown: One of the issues is, then, how do we discover borders. Alexandru Petrescu: We tried to use ULAs in vehicular networks, within a vehicle. There are a few practical issues: what is a site, what is the scope of computation of this ULA. In homent, it is simpler, I think, within this house or that house - but is it just within a house, or within an apartment block, .... How to form these ULAs, the RFCs give several options, need a recommendation as to which to use. Tim Chown: Issue on stability of ULAs. If you want stable communication you presumably want stable ULAs per link, and the same ULA generated each time. Alexandru Petrescu: Should house-to-house communication use ULAs? Tim Chown: No, should be global. Michael Richardson: We have to fix the host selection issue. There will be ULAs, and there will be devices not taking a global address and they will continue to use them. There will be people with multiple upstreams, that will need the host selection anyways. I believe strongly in happy eyeballs and that it will fix everything. No, I think that there will be problems, but that they will be fixed. Anders Brandt: What are the use-cases? Reminder, sleeping devices, like a light-switch, will not be awake to see any new prefix delegation, and you want more static configurations. Lorenzo Colitti: I believe that current implementations should work today, but it hasn't been tested large-scale, though. (Naming and SD slide 6) Tim Chown: Probably the part that was attracting the most Kerry Lin: We had a BOF in Atlanta, on this subject, and we are very much interested in solving this problem; Homenet is in scope, we're discussing on mdnext mailing list, and we will have another BOF in Berlin, with the intent of having a WG. Tim Chown: Wraps up, a -08 is in the work, we probably will need another WGLC when the text is done Ray Bellis (Chair): Probably be a full-length WGLC, given the amount of changes. Lee Howard: Still very active comments on the list, so give it time, no rush of a -08. Mark Townsley (Chair): Wants to see WGLC done before Berlin. Take on no more documents as WG documents, until this doc is through WGLC 9:34AM-9:57AM - IEEE 802.1Q for Homenet - Paul Unbehegen Presenting the scope of 802.1Q work, basic principles of Ethernet, .... Difference in Q is that there's no "spanning tree", no "unused links", using IS-IS to create a SPT. Guaranteeing that packets take the shortest path, and no links are ignored, all are visible. Supports costs etc. Ted Lemon: Are you encouraging that we deploy this in homenets Bob Hinden: One thing you said that resonated with me was, that routing protocols do not take a lot of CPU and can be run on small devices; important data point for this wg. Ole Trøan: Looking at this from L3/Homenet, where I want to use a routing protocol, do you see any reason or use for using the same routing protocol on L2 and L3? Paul Unbehegen: YES; the same state machine that sees the entire topology, that can lead you to make some very nice heuristics. Peter Lothberg: Comment...when you were talking about....there were different ways, routing protocols, spanning trees, ... What we learned from the big internet is that it runs fine as long as you run "ships in the night" - when you start to copy from one to another, we shoot ourselves in the head. Either we have different domains, ships in the night - or we specify something. Mark Townsley: You're using this at L2 and L3, you can create larger and larger bridge domains, and what have you. Would you run your home as one gigantic bridge, or? Paul Unbehegen: Depends on the home, I guess. In my home, I would prefer to have both. Lorenzo Colitti: I bought a ... switch, and it didn't support spanning three. Cost 25$. The model higher had a light that comes on when there's a loop. How does it work when you stick a stupid device in the network? Paul Unbehegen: Many things you can do when you see a frame, blocking, .... 9:57AM-10:10AM - Chris Grundemann - Hipnet General presentation of the features and status. Believes that Hipnet meets all principles from the homenet architecture, although makes the point that it is a little hard to extract the principles (reference to Tim's presentation with explicit enumeration hereof, or not) Michael Richardson: Correct me if I am wrong, I Read the I-D very quickly....I do not think that you support more than 1 upstream ISP.... Chris Grundemann: The default method is active standby failover, but there is a mechanism for having both active at the same time. Paul Unbehegen: Sounds like you are trying to build a spanning tree. How do you do load balancing Chris Grundemann: Building a hierarchy, choosing 1 path to the internet, 1 uplink port, using the others as LAN ports. IT is an addressing scheme, using hierarchical routing over it; can run a routing protocol on top of this. A very very simple approach without running a routing protocol. Using existing protocols with a little tweaking. Paul Unbehegen: May actually be easier to use IS-IS Chris Grundemann: IS-IS is not doing prefix delegation Kerry Lin: Homenet I-D fairly silent on multicast. Acee Lindem: This is going to cover some subset of the homenet topologies, but likely not all of them. Not sure if it handles the source ingress filtering problem, the complete generalized multihoming solution. The question is if it handles sufficient cases to be a solution. Lorenzo Colitti: How does this react to topology changes? I have this historic hatred to DHCP, as DHCP can give but not take away. When you plug and unplug things, even if you use lots of tricks, you still seem to run out of addresses as you give addresses out but can't take it away. Does it deal with unplugging? Chris Grundemann: In the testbed we haven't seen it, but we haven't tested it exhaustively. Lorenzo Colitti: My point is, that you can't retract the leases you've given, except if you're no longer up. If a second... Mark Townsley (Chair): Let's take this off-line. Appreciate this work, while new to the list it's fairly well advanced, it seems. 10:10AM-10:26AM - Fred Baker - IS-IS/OSPFv3 Extensions for dest+ Motivation and explaining: one question I got from everybody was, "this is a multi-topology problem, why aren't you solving it as such?" - doesn't believe that it is: it is a reachability problem, a qualified route to a destination, and not a topology problem, and so approached it as such. Walk through how prefix distribution was added to IS-IS, copied from OSPF - easy draft, fixing IS-IS. Spoken to OSPF WG chairs, there was a draft 5-6 years ago on making extensible LSAs for OSPF - Fred is in support of that, and it is likely to be resurrected and moved forward. Will revamp the I-Ds accordingly, if they have a head start towards consensus on this extensible-LSA format. Peter Lothberg: I have some extra work that you can do, Fred, Could you extend ES-to-IS? We are starting to carry policy, can you make it such that the end system can tell the network what they want to do? Fred Baker: One of the complaints I have with src address selection is, that it essentially makes the ES make routing decision. Peter Lothberg: I think that what we are trying to do is build a dist. database? Fred Baker: Already have that, called the routing table Peter Lothberg: Doesn't extend to the end-system, could we make this happen? 10:26AM-11:02AM - Ole Trøan & Lorenzo - Source and Dest. Routing Walking through the slides Acee Lindem: Another thing that is implicit is to use the RP for prefix delegation. Whatever you do, I believe that you want to do it in a single routing protocol instance. Ole Trøan: Haven't talked about how to do it in multi-topology, didn't consider that. Rajiv: Insist on that things must work, be simple, and suspects that there are lots of loops that can happen, and it shouldn't be hard to debug. Lorenzo Colitti: When Ole said "micro-loops", he meant convergence. Ole Trøan: A set of problems we're trying to solve o Policy based routing o Multihoming o ... Teco Boot: I am interested in solving the legacy routing problem, and I am using tunnels: if a router doesn't support this new mechanism, then it can tunnel to the egress router? I do not think that this is really a "hack" (reference to slide 7) Acee Lindem: This, you call it "hack", it is no worse than like the edge detection or active stand-by in hipnet. If you are talking about interim solutions, this actually could be one. It's really a more generalized heuristic than some of the hipnet points. Rajiv: Referencing slide 5 See the intend, and somewhat agree with preferring no 1. But....there are going to be cases where the destination isn't living in the ISP or beyond that, but in the homenet and then we will have an issue Ole Trøan, Lorenzo Colitti: If the routing protocol is broken, then yes, we can't get to the destination Rajiv: Difference between being blackholed and being misrouted. Dave Thaler: The way I looked at this, the *,D class is for things within the home, the other is for routing to the outside world where the assumption is that the outside world has ingress filtering, whereas inside the home there is no ingress filtering. Should make that very explicit! Rajiv: Have you considered something like L3VPN...? Lorenzo Colitti: We welcome your proposal to do so. Alexandru Petrescu: Relate to prior work done elsewhere, in WGs related to mobility, there are some existing specifications that are a little more general, in ProxyMobileIP The other comment is related to key lookup in the fib, use two keys. I think that you said that the search in the dest is using longest prefix match, source also - I am not sure whether that is the good thing. Dest, ok - but source, should it not be exact match? Lorenzo Colitti: No, otherwise you would have to populate /128 routes Ole Trøan: Do not have a reliable way to know which addresses are in use, a host may autoconfigure it, without the routing system knowing. Alexandru Petrescu: Use DHCP or SLAAC, or? Lorenzo Colitti: Any mechanism you want, statically, .... Alexandru Petrescu: Microloops w/ ingress filtering. Maybe it should not be or is deployed in home. I am not suggesting that it should be present in each router in the home....but if it is deactivated, then you get this triangular routing..... Lorenzo Colitti: In home it is shortest path, outside it is src,dest So in a home, triangular routes may occur - but that is not a problem. Lee Howard (but, didn't state his name at mike so not sure): You have internet & two ISPs (slide 3), how do you make the SP tell the home that "this is the set of destinations I provide connectivity to"? How does a service provider install a more specific route into the routing table? Ole Trøan: How do we get the default route into the home? RFC6204 has the "more specific route" option, inject those using the RA option; will be injected into the homenet a AS externals. Pascal Thubert: Work to extend 4991 planned? Ole Trøan: Not currently. Presenting Slide 8, ask question to proper behavior when an ISP connection goes down. Teco Boot: Yes the host should notice if a link goes down. Was some discussion on that on the mailing list. Lorenzo Colitti: The only signaling mechanism we have, dynamically, to signal to the host is to change the lifetime on the prefix information. As for 64, for 3G backup, we do not support that at all. Pascale Thubert: Back to the previous discussion. You will find more and more devices for which renumbering may be costly: sleeping when it happens, etc. The answer may be that we can't do it today, but we can do it tomorrow. We do have the PIO, so we could indicate that this prefix is preferred, .... Ole Trøan: Let us be clear, we are not renumbering, we are just setting the preferred lifetime to zero, not yanking the prefix away. Andrew McGregor: Deprecating addresses when a link goes away is the right thing. Because, if you don't, if the link comes back and doesn't get the same delegation again, then you squat your addresses. Ole Trøan: OTOH, you do enter a contract with your provider, who promises "these addresses are valid for 3 months"... Andrew McGregor: Well, sometimes they may not be able to do that, operational issues, .... Rajiv: Agree Jari Arkko: Think that this is the right approach. A little cautious....though....if a little glitch in the network happens, in the ISP connection, we should delay and buffer.....in favor of keeping the last source prefix around. Lorenzo Colitti: We have not put enough thoughts into that Eric Vyncke: How do you detect the ISP connection going down? If the CPE is up, it can, if it is the CPE that goes down, how would you do? Ole Trøan: OSPF allows for that Eric Vyncke: And so you remember from whom you get the prefix... Lorenzo Colitti: If you do not know anything, don't deprecate Dave Thaler: Yes, deprecate Yes, dampering is fine No, do not avoid deprecating the last address. Chairs polling who has read this document - some, but not enough. Request more people reviewing the document, and encourages that a -01 be published with what was learned today. 11:02AM - Ole Trøan - Homenet Implementation Report Mark sets stage, follow-up from last IETF where IPSO lent space; generally favorable from WG for more implementation reports, so here it is. Continuation of the presentation Jari gave at the last IETF, alas, with Jari now having a job requiring him to wear a jacket and a buttoned-shirt, he's no longer able to hack [as much] during an IETF week, so this befalls on Ole. Reporting on a group of folks being sequestered in a villa during this IETF, not allowed to leave except to report on the code they're developing. Showing pretty pictures with gear and champaign.....shouldn't pity them too much Walking though slides ; experience that the service discovery bit was a little complicated and requires a bit more thought and work. Ralph Droms: Back up two slides, to service discovery. Do you have any sense if your experiences are specific to your design/implementation strategy, or if it applies to everything regarding mdns Markus Stenberg: mDNS has about 200 should/must, we implemented them, tested with Apple. Your typical mDNS creates loads of stuff, how many mDNS records you can fit within a single LSA, even when compressing: answer, one Apple-devices output doesn't fit inside one LSA. This might, thus, be painful and might not be to recommend. Conflict resolution...the only way to play is to not play at all: detect, and then drop. Michael Richardson I think that Marcus answered the question: you proxy into OSPF and distribute it using OSPF, correct? Ole Trøan: Yes Jan Žorž: Were all these addresses on the same or on different interfaces? Ole Trøan: Yes Jan Žorž: So, you selected the source, which in turn selected the gateway, interesting, this is the solution to what I was screaming about in Paris Ted Lemon: AD hat off. What I just heard at the mike about mDNS was that when it breaks, you just drop it. We need to fix that? Ole Trøan: Yes, very much so, need to sit down and do that. Jari Arkko: Thank you for doing this, it is great, and good to see this go forward. I want to apologize for not participating, but I also want to say that we've been doing some stuff, hired a student - it's just the reporting that has been a little deficient. Closing comments from chair: Need to see some progress before Berlin, LC for the arc document, and decision on routing protocol or no-routing protocol, etc .... Michael Richardson comments that this progress is too slow.