***************************** **** Preliminary minutes **** ***************************** =========================== IETF 67 AUTOCONF WG Minutes =========================== Minutes taken by: Ryuji Wakikawa Emmanuel Baccelli Simone Ruffino (merged & edited by Thomas Clausen) Links & References ------------------ Jabber log from the meeting (Scribe Justin Dean): http://www.ietf.org/meetings/ietf-logs/autoconf/2006-11-06.html Proposed MANET Architecture Model (presentation: Thomas Clausen) ---------------------------------------------------------------- Background: draft-chakeres-chakeres-manet-arch-01 was first approach at making explicit implicit assumptions in manets. This presentation aims at presenting an architectural model, complementing this I-D. Slide: Misperceptions - assigning a subnet prefix to wireless interface and assuming that all MANET routers are in the same subnet causes a lot of problems. For example, if configuring a MANET in this way causes ICMP redirects, and as illustrated in draft-iab-multilink-subnet-issues-01, this also breaks assumptions both for platform- and host applications Joe Macker (JoeM): - people *have* been configuring MANETs this way, but it is a good thing to make explicit that ththis is not a good thing Slide: Architectural Model - First, here's a router (R) with a manet interface (top) and a host (h) (bottom). (animation) - Of course, a router may have multiple hosts connected, over an interface of a well-known type such as an Ethernet. Or, indeed, many such hosts - In MANETs, the "node" is often thought of as encompasing a "host" and a "router" in one box - A MANET router, which is given a prefix, assigns this to the non-manet link(s) - the manet interface is configured with a /32 or /128 (i.e. subnet unique) adress, is unnumbered (v4) or has a link- local address (LLA). - Autoconf tasks, then, are to assign LLA and unique prefixes. (for more details, please consult slides) Thomas Nartan (ThomasN): - Having a dotted line around the router/host would help conceptualize that this can be one box - The manet interface is, then, what we're interested in. The hosts see only a regular well-know link type on the bottom-part Dave Thaler (DaveT): - if those devices are in all host then if the link on top has some wierdness then the applications don't see the wierdness because of this abstraction sepeartion of router/host on one node Fred Templin (FredT), ThomasN: - Let us call the bottom "normal part" of the model the classical IP link model? Jari Akko (JariA): - We are dealing with normal router, and normal hosts? Thomas Clausen (ThomasC): - Yes, except for the "weird" multi-hopn MANET interface seen only by MANET routers. ThomasN, DaveT, JoeM: - see it as a logical box router+interface. ThomasN: - can hosts interface be wireless ad hoc? Bob Hinden (BobH): - Yes. This is only an abstract model, this is not an implementation solution. DaveT: - Yes we must not mistake the abstract model issues and the implementation issues. ThomasN: - if everything is as normal, then what is the problem with autoconf? DAD? ThomasC: - the MANET router interface configuration is the problem that needs to be solved. JariA: - the dynamics of the MANET router topology is a problem. Charles Perkins (CharlieP) - Do you intend to impose this model for simple MANET nodes like we have envisionned in MANET? ThomasC: - Yes CharlieP - We need to allow a solution that takes into account nodes that pop in with an address already configured. JariA: - that is a reasonnable requirement. DaveT: - We should use another term than DAD. It is already pointing at a specific solution. ThomasC: - Is there a conscensus on the fact that this model is reasonnable as a basis? JoeM: - We need to write it down. CharlieP: - I still don't understand: How does it work when there are Hosts on the MANET interface? ThomasC: - There are only MANET routers connected to MANET interfaces. This is the logical model. Emmanuel Baccelli (EmmanuelB): - the very first slide answers your question. BobH: - I agree with this model. It is good enough. The best is the enemy of the good. JariA: - this is a good model. DaveT: - The R box needs to also feature "DHCP or DHCP relay"-like functionalities to serve the hosts (if any). This is in addition to the autoconf tasks we have talked about so far. Or rather: the "description" of the R-box should not exclude that it may supply "other configuration information" to the hosts which are attached to it. ThomasC: - Yes, but this is a little further ahead. We will solve this problem in a second phase. DaveT: - OK, but don't design something that is just for addresses. We should have something that is extensible for some more information. JoeM: - Yes, we will capture that. ChalieP: - I disagree with Dave Thaler. Even in IPv6 we do not do that. This should not be a requirement to do a general autoconf, instead of the initial goal: just addresses. Justin Dean (JustinD): - I guess we all agree then: the first phase is definitely address autoconf. CharlieP: - DHCP has a single point of failure... ThomasC: - We're sliding off topic here. I think that the idea is to state the desired functionalities, but not enter the solution space yet. DHCP is one way, may or may not be the way to go, but that's a solution-space issue. Simone Ruffino (SimoneR): - Please put examples of assignments. Numerical examples of addresses. ThomasC: - We will discuss that on the mailing list. ThomasC: - Can we hum in favor / disfavor of this model *** Humming in favor, none in disfavor - we move forward with *** *** documenting this and discussing on the mailing-list. *** AUTOCONF Problem Statement (presentation: Emmanuel Baccelli) ------------------------------------------------------------ Using dhcp for Autoconf in MANET (presentation: Fred Templin) ------------------------------------------------------------- - Presentation of how the OSI layer 3 sublayer models may fit MANETs draft-templin-manet-autoconf-link-00.txt BobH: - Not sure that the OSI model was designed with MANETs in mind, not sure how well it fits FredT: - If you are interested, please do read the draft draft-templin-manet-autoconf-link-00.txt ThomasC: - is this contreversial regarding the previous discussion on the architecture model? FredT: - this is more about background, where solution spaces may grow ThomasC: - Are you talking about a prefix on the MANET interface? We were talking about configuring an address. FredT: - As long as a prefix is used by only one mobile node. It does not create problems. DaveT: - That's true enough. - Presentation of how to make DHCP work over MANET draft-templin-autoconf-dhcp-02.tx - Works with IPv6, not so well with v4 JoeM: -I had more problems with DHCPv6 than with v4, contrary to what you state? FredT: - We should talk about this offline.