Draft MEETING MINUTES ===================== Ad-Hoc Network Autoconfiguration (autoconf) Agenda Meeting : IETF 73 Tuesday, November 18th 2008 Time : 1520-1700 Location : Salon G Chairs : Thomas Heide Clausen Shubhranshu Singh Jabber : autoconf@jabber.ietf.org Audiocast: http://videolab.uoregon.edu/events/ietf/ URL : http://www.ietf.org/html.charters/autoconf-charter.html http://tools.ietf.org/wg/autoconf/ https://datatracker.ietf.org/public/meeting_materials.cgi?meeting_num=71 ================================================ Note Takers: Charles E. Perkins, Emmanuel Baccelli ================================================ Shubhranshu: Agenda bashing & Status update * Mobile Ad hoc Network Architectural Discussion http://www.ietf.org/internet-drafts/draft-clausen-manet-linktype-00.txt (also refer to draft-ietf-autoconf-manetarch-07.txt) Thomas Clausen (TC): - We had achieved some rough consensus as of IETF 67 - Dec.07 -- publication requested... - draft-iab-ip-model-evolution-01 is influential - Content: "node taxonomy" & "address architecture" - Node Taxonomy: nodes have two "parts": a classical part and a "manet" part. - discussion about link asymmetry - Dave Thaler: refers to a definition of "link" to be the object across which TTL=1 makes sense Dave Thaler: The term link means different things in different communities. It may mean a L2 segment in some cases, in other cases it may mean a single layer 3 "hop". Charlie Perkins: Agreed. So we must clarify this in the document. It may be counter-intuitive the way it is now. E. Nordmark: If we are going to use multi-link subnets, we must be very counscious about it. - Erik Nordmark: If we are going to have multi-link subnets, it ought to be a very conscious decision. - Manet: "neighborcast" - Link-local multicast/broadcast vs. "neighborcast" Charlie Perkins: The use of the term neighborcast is maybe confusing here. It is different from multicast. T. Clausen: We are just trying to characterize what is the subset of nodes that receive a transmission without L3 forwarding. - Dave Thaler: There is a confusion between "Link" and "link", where "Link" has more to do with TTL=1. - Jari Arkko: Can construct a mechanism that hides the neighborcasting. - Thomas Clausen: Then we have things like 802.11s - Jari Arkko: we are not here to do layer-2 things - Thomas Narten: neighborcast is a layer-2 property - Dave Thaler: What's the subnet topology for the link-local prefix? Does link-local multicast escape the boundary of neighborcast? - Seung: Wouldn't it be better to pick a new term instead of "link"? - Thomas Narten (TN): Confused about which layer we are discussing. - Thomas Clausen: We are considering Manet nodes doing Manet routing. - TC: shows picture of classic link model subnets attached to manet cloud routers - TN: Need to explain how the the different interfaces are handled - Teco Boot: We are using these link layers a lot in some environments. The document should be discussing what the link layer is providing to us. - Zack Shelby: 6lowpan already has the architecture and the problem statement and some solutions. The "Link" definition will (has) cause a lot of pain. Shrink the "Link" down to a "link". - TC: I may have a link with 1-hop connectivity. We are used to thinking about links. The document is supposed to document the experiences we have had while building manets. - Zack: The mistake is trying to model this like OSPF? Why change what works? - TC: Maybe we should take it to the mailing list? - Dow Street: Using a few words to mean a lot of different things. To most people, manet and link mean the sort of wireless objects we are used to. But then the document claims that ethernet also satisfies the property of manet link. - TC: The document is descriptive, not prescriptive. Ethernet satisfies the properties, and _more_. - Dow: Seems like you are saying that a Manet link is one over which you run Manet protocols. - Erik Nordmark: [roll] chair explained how IP is used over the radio links. If you want the nodes to be plug-n-play, they have to find each other. They need to find nodes that are one hop away. If you make a link be one radio hop away, then some subnet things don't work. - discussion about mobility breaking "bridged" (or is it "not-bridged") structure for subnet - TC: My node's prefix wouldn't change if move from place to place - Erik: Are you using the nemo model? - Erik: Does manet use only host routes - Joe Macker: No, manet DOES allow use of subnet prefix - Erik: Do you need to have a subnet prefix to every router? - Erik: Or assign a /128 to every router? - Erik: Can't use a single-hop protocol to reach a node several hops away. - Erik: Problems with address delegation - Joe Macker: we do this now (??) - Erik: wants to understand the picture - TC: Assigning /128 in some situations. - Dave Thaler: It's like nemo, when you have hosts stably attached to a manet router. - Jari Arkko: The analogy to nemo is not very useful since the techniques are not similar. - Joe Macker: (1) Would like to see link-local mean radio range - various disagreement about what people said link-local should mean - Erik: to reach farther than radio range, have to use subnet definition (presumably with multi-link subnet). - Joe Macker: virtual link type, manet link type - Dave Thaler: this is a semi-broadcast link Type, NBMA, - Joe Macker: it's OK to talk about link characteristics. Some people think there is one "kind" of radio link. IP layer actually manages the links quite differently. - Joe Macker: should bounce some of this off the routing community - Thomas Narten: Links are defined by radio range. Point to point links work. The working group doesn't seem to be able to agree on anything. It would be a big step forward to agree on one model. Have to be precise on terminology. - Seung: Regarding the comment about point-to-point links, this is exactly how they are building manets, and it seems to work fine - Zack: Not complaining about the set of circle links. Propose to call a link one radio range. 6lowpan is doing this. The group is doing IPv6 and autoconf for IPv6 networks. - TC: You are preaching to the chorus here, but that approach has been criticized in the past. - Zack: You would have support from the 6lowpan working group. Also have support from the other AD. - Jari Arkko: The issue has not really been to use one definition or the other. It's been how to describe things so things still work. Wants to see the document that explains these things. - Charles Perkins: Strongly support the link == radio range Also subnet == TTL=1 - Erik Nordmark: Thinks the subnet proposal is wrong. Also, with proposed definition of link, link-local concept is not useful for applications. - TC: don't necessarily want subnet prefixes - Erik Nordmark: wants prefixes to allow movement without changing IP address - TC: But the manet infrastructure is allowing you to move without regard to the subnet prefix - Erik Nordmark: But then it is spanning the whole cloud - TC: But we have this mobility today without the same prefix - Dave Thaler: using the linear arrangement of circles. How many links is N3 attached to? He says 3. How many interfaces? 3 RFC 2461 says an interface is attached to a single link. (need to disagree with this) Radios are unidirectional??? (need to disagree with this) - Shubhranshu: hear opinion on radio range link and none against documenting the radio link cocept of link - TC: Does not think it is possible to build consensus on the document and properties of radio links - Jari Arkko: It is very important to explain the architecture It will help understand how manet fits the Internet architecture. - Shubhranshu: The document has good merit and need to elaborate on some missing components, as pointed out. - Erik Nordmark: In the picture, the link centered on N3 contains N2 and N4. C. Perkins: I propose that a link is a radio range and that a subnet is what can be reached by packets with TTL=1. C. Perkins: I think this document is a good start. I think there is an almost consensus on the fact that a MANET link is the radio range. Thomas Clausen: I disagree. There is no consensus. There are many different opinions. The only consensus is that we can build stuff that works. But there is no consensus on the model. What is needed more to describe the model? - Joe Macker: Are there any 6lowpan documents that describe these concepts in a way that is agreeable? Shubhranshu: Running out of time, plz take further discussions to ML