Autoconf WG (INT Area) Wednesday 12th July 2006 Agenda Bashing, Status Architecture document (Joe Macker) Manet and Autoconf WG chairs are co-editing (lead Ian Chakeres). Intention to publish -00 soon (1-2 weeks) after IETF. Challenges, design considerations, taxonomies, scenario examples. Draft outline. Dave Thaler: don't see discussion of subnet model, where does it fit in? Previous document is intended as input. Jari Arkko: Agreed, need to have architecture model can develop protocols for. Dave Thaler: is there one architecture or multiple architectures? The subnet is the architecture. Jari Arkko: Should pick one architecture. Fred Templin: Should define "link". Charles Perkins: How much of document taken up on subnet issues? There are other important issues. Problem statement (Emmanuel Baccelli) Converging three documents (ADP, framework, scenario) into one. Table of contents. Constraints: wireless, pseudo-broadcast, dynamic topology, relaying. Disconnected, one gateway (stub), multiple gateways (multi-homed), gateways may be mobile. Joe Macker: Suggest alternative terminology, stub not right word for one gateway. Fred Templin: Multiple gateways may not mean multi-homed. Framework is state machine. Nodes can start in different states. Mechanisms for transition between states. Pre-service and in-service DAD. Charles Perkins: Node starting may not be able to tell if network is connected or not. Also nodes should be able to keep addresses they already have. Joe Macker: Solutions may not involve DAD. Not used to problem statement including part of solution. Possibly may not include DAD at all (e.g. trust unique MAC address) so shouldn't be requirement. Need place in state machine to turn off. Dave Thaler: Apparent overlap between documents, should be resolved. Should not have any state machine in problem statement. May not be a concept of tentative address. Thomas Clausen: Don't disagree with what's said. Look at state machine as general model. Must not mandate DAD, tentative address, or may indicate tentative address automatically becomes non-tentative. Christopher Dearlove: State machine looks too formal, concepts. Kenichi Mase: Reference model, not requirement. Joe Macker: Arrows exist that don't see. Problem with this in document. Goals and requirements: These are for solutions. Joe Macker: Continuing to see solutions, which should not be in problem statement. Framework (Kenichi Mase) Update of previously presented ID. For stand-alone MANETs. Removed redundant descriptions, added duplicate address announcement (DAA). Joe Macker: Developing SMF in MANET, might be used rather than control plane. Fred Templin: These are necessary mechanisms in stand-alone MANETs, may not be needed with gateway. Thomas Clausen: Should not say "each node should perform" but "protocol suite should provide solution, which node may or may not use". Robert Jackson: What overhead does this reduce? Joe Macker: This has been done in OLSR, but problem is that this makes it part of the protocol. Why mentions SMF, may be protocol independent. This says framework, but says how to do things. Thomas Clausen: Interesting document, but needs to be separated and not solution. Interest in OLSRv2 integrated with this idea, but in this WG need to be more general. Gateway Aggregation Protocol (GAP) (Kenichi Mase) MANET attached to Internet via multiple gateways. Common prefix advertisement by gateways, avoids DAD. Dave Thaler: What subnet model does this solution assume? Slides only say address in singular. Simone Ruffino: Uses terminology which differs from problem statement. Edge mobile node here is Internet gateway in problem statement. Solution appears complicated. Joe Macker: Not a use case used to, interesting but specialised. Where (in which document) should we discuss multiple gateways? Yangwoo Ko: About terminology: called gateway aggregation, but as used here, only have one gateway. Simone Ruffino: Signalling between access routers is not trivial. Autoconf for MANETs with Multiple Gateways (AMG) (Simone Ruffino) Aimed as general purpose stateless solution, particularly with multiple gateways. Global addresses flooded from gateways using different prefixes, ranking algorithm for selection. Avoids DAD. Charles Perkins: Question about possibly frequently changing global address. Need some hysteresis mechanism. Dave Thaler: Does this assume single subnet model? Yes. Use ULA as MANET local address. Use SMF for prefix announcement, routing protocol advertises addresses. Include validity time. Fred Templin: Why not use routing protocol to advertise prefixes? May be more efficient. Answer, want RP independence, increased overhead not large. Joe Macker: If design TLVs for this process to use in SMF, could easily apply same to routing protocol, flexible. Thomas Clausen: Suggest design specified in terms of packetbb. Charles Perkins: What happens if node has existing address, joins MANET, gets second address. Can we do this without configuring ULA? Fred Templin: Suggest using MANET local address to operate routing protocol. Charles Perkins: May instead want to use existing address. Do not see why automatically need ULA. Fred Templin: Need to have joined MANET to have global address. Charles Perkins: May have scenario was connected to Internet, then disconnect and join MANET. John Schnizlein: How to know closest? Joe Macker: Can count hops in MANET. Christopher Dearlove: Also have hop count in packetbb/SMF. Also may end up using gateway for routing other than that providing prefix. Agreed to take offline and to list. Implemented for OLSRv1, considering OLSRv2 and DYMO (proactive protocol is easiest). Dave Thaler: The problem statement document should make it clear that a host can have as many addresses as it wants. Charles Perkins: Hosts should not have to get addresses they do not need. MANET Autoconf using DHCP (Fred Templin) MANETs attached to Internet via one or more gateways. MANET at L3 needs to use DHCP may be multiple L3 hops away. Use MANET local address for routing protocol. Node is DHCP client for global address. Details of how to implement this using tunneling and other mechanisms. Simone Ruffino: How to choose new gateway? Based on similar mechanisms to previous presentation. Compatible with "NETLMM using DHCP". IPv4 and IPv6. Can route out using any gateway. Downlink uses preferred gateway. Simone Ruffino: Most traffic will be downlink. Does this approach produce efficient choices? Answer is there is a gap to look at. NETLMM using DHCP (Fred Templin) Discussing what happens in backhaul network, outside MANET. Nodes may move within a domain, even to different MANET, keeping same address. Uses mobility anchor point (MAP) associated with DHCP server (not necessarily co-located). Also tunnelling in backhaul network. Ongoing work in NETLMM WG. Final comments Joe Macker: Difficulty with getting documents out next week is coordinating contents. Jari Arkko: Need to get documents out. Should discuss on list. Thomas Clausen: Need to get documents out. More may be added. Joe Macker: Issues with the problem statement document have been raised, and need to be taken on board. Thomas Clausen: Now is time to write up and present solutions. Documents need to cover real needs.