Updated 4/5/05
As part of the IEEE Communications Society's INFOCOM conference, on March 17th, 2005, a panel discussion was held to contemplate Internet connectivity for mobile devices and users. This topic has been of significant interest to both the academic and commercial research communities. Numerious differing mobility technologies are evolving in labs around the world, and some are being actively developed by standards organizations and/or for use in products. This panel discussed the current state of a number of individual technologies and the arms race between them.
The panel was organized by Wes Eddy, moderated by Joe Ishac, and featured Will Ivancic, David Maltz, and Pekka Nikander as speakers.
Presentation slides for all the speakers are available.
Additionally, the panel abstract and proposal is available, although there were several personel changes between the initial proposal and actual panel, due to scheduling and travel conflicts.
This webpage aims to document some of the important points that were made in the panel discussion, and provide links to relevant papers and websites. This should help to foster productive research in the areas that the panelists identified as open areas.
The introduction reviews the design of the Internet protocol stack, and how layers communicate with each other. We talked about how mobility, unlike some other services, can be provided at several different layers, without any clear best layer to put it in.
The key was that there are many ways to do mobility in the Internet, but that Mobile IP has been "first out of the gate", and we want to know if multiple mobility protocols can coexist in the Internet. Specifically, we'd like to compare different schemes and see what's best for various entities (users, ISPs, etc) in terms of cost, deployment, ease of use, security, extensibility, scalability, etc. Since this is a very large question, there is no way we could fully cover it in only an hour and a half, and there is still no clear answer. Hopefully this webpage can help the discussion continue in various groups.
Will's material focused on mobile IP (v4 and v6) and some of the practical issues with deploying the protocols in a secure environment. Will also talked about network mobility (as opposed to host mobility).
One of Will's main points was that security and mobility solutions seem to be much easier within an administrative domain than they are across domains. Improving the inter-domain difficulties could be one area of improvement for these protocols.
A useful paper is Securing Mobile Networks in an Operational Setting, by William D. Ivancic, David Stewart, Terry L. Bell, Phillip E. Paulsen, and Dan Shell, from the IEEE Computer Communications Workshop 2003.
Will also briefly mentioned ad-hoc networking, which the IETF's MANET WG is standardizing. Specifically, Will noted that radio technology that is specially designed for ad-hoc networking would be a key improvement.
Pekka's material focused on the Host Identity Protocol (HIP), which aims to introduce a separation between a host's identity and its location.HIP uses cryptographic techniques to securely achieve this, and can also be used to provide advanced mobility and multihoming services.
One useful paper is Integrating Security, Mobility, and Multi-Homing in a HIP Way, by Pekka Nikander, Jukka Ylitalo, and Jorma Wall, in the Proceedings of Network and Distributed Systems Security Symposium (NDSS'03), 2003.
David's presentation focused on session layer mobility, and MSOCKS. These techniques fit into the stack at a higher layer than Mobile IP, or HIP, and involve different pros and cons, which were well-explored in his presentation.
A good description of the MSOCKS system is in the paper: MSOCKS: An Architecture for Transport Layer Mobility, by David A. Maltz and Pravin Bhagwat, from IEEE INFOCOM 1998.
Session layer mobility is discussed fairly well in the papers on the Migrate Internet Mobility Project webpage.
For one view on "application layer mobility, see: Application-Layer Mobility Using SIP by Henning Schulzrinne and Elin Wedlund, from ACM SIGMOBILE Mobile Computing and Communications Review, 4 (3), July 2000.
Questions appear in bold. The text here comes from notes taken from a somewhat noisy recording and is only a rough approximation of each speaker's actual words. Please don't hold them to any of this!
Joe asked the panelists about security with regard to rogue nodes
Will noted that he brought this up in his talk, about protecting mobile routers from hijacked interfaces. He said that crypto-based solutions will have some scalability questions due to key management.
Pekka said that there are 2 sub-problems here: 1) do you get your packets through or not, and 2) policies and payments - there should be integration between payment solution and access point ID.
David answered that access points not forwarding is an easily detectable DoS attack, and less serious than a hidden access point generating false billing statements on your behalf. End-to-end encryption can protect user data, but the real solution is to use multiple interfaces or link technologies, so that a trusted link can usually be found.
From the audience: What is the best solution for seamless mobility, without delays and losses during handovers? For example, when driving in a car
David said that this is the holy grail - that all should be seamless, but reality is otherwise.
Pekka noted several dimensions that have to be considered here. Whether you are considering a mobile host or a mobile router is one. Another is timescale - how often do you make handovers - per second or per day. And also, the topological span of the handover is very important - seamlessness is only achievable within a single access network, but in this case, even with very frequent handovers. Layers 2 or the lower part of layer 3 are good places for frequent handovers, but this all depends on your definition of seamless.
David said that that's one point for having multiple solutions! - not everyone's demands are the same.
Will mentioned that when you change technologies, sometimes you have to interact with the network, for example typing in a password is required at times. A smart card with credentials to get on multiple service providers might be a lot better.
From the audience: I have 2 comments and a question. 1. Triangle routing problem that everyone mentioned is solvable with multiple home agents. 2. There is a connectivity problem in addition to the addressability problem. Q: According the the end-to-end architecture, how are these solutions going to interact?
Will: HIP is intriguing as a tool, getting away from the address-based identity is a necessity.
Pekka had some comments on the comments! He said that if you're going to have multiple home addresses, just forget about mobile IP, because that's not the point, and watch out for a routing table explosion problem. To the second remark about layering, from the e2e principle, mobility should stay at the lower layers to avoid having to replicate it at all upper layrs. HIP is OPTIMAL at layer 3.5.
David agreed that HIP is at a sweet spot in the stack. It saves from all the transports requiring changes; but we will have solutions across layers, we won't be able to have a single solution.
Pekka noted that congestion control might need some work. There's a difference between doing it on single hop and multiple hop paths, and minimal signalling between layers is desirable.
From audience: Indirection, location management, and signalling : how do you put those together?
David said that this division is an accurate assessment fo the problem, and he noted HTTP work.
Pekka said they've decided to leave indirection and location management out of the HIP core, there are multiple ways to do these. DNS, distributed hash tables (as in Hi3 - HIP + i3 as rendezvous).
(Garbled) Something about businesses and controlling traffic
Pekka has some comments on business models. Mobility is for the users to implement with minimum infrastructure, you don't need any business model, users will come, which is probably not good for my employer (who sells infrastructure equipment)!
David thinks that there is an incentive for operators to provide good service to keep you on their network. As technologists, we don't want to make it too easy for them to lock us in, and prevent us from roaming across networks.
Will is using a lot of networks in ways that people might not like to know. For example, mobile networks using one IP address, which is fine for a single researcher to do, but when more people start too, it could swamp or stress the network, and charging plans would have to change.
From the audience: does HIP require changes to congestion control?
Pekka said that it does to some degree. TCP over multi-homed HIP will need work, but not right away, SCTP will need to be reworked a bit. Right now HIP is only specified to use ESP, no matter what the transport is, but that's in the process of being relaxed, which is bringing up demuxing problems that will need to be solved.
Someone: How does HIP solve reordering?
Pekka: that's someone else's problem.
Will: What work is being done to take HIP to multicast or NEMO functionality?
Pekka: multicast is not being done directly, Hi3 work on application layer multicast is underway, as for NEMO - Pekka has a PhD student working on it, that is not yet published.
Joe asked about PKI, are they essential or is persistence of who you're talking to more important than they're exact identity?
Pekka: that's an application layer problem
Then it was TIME FOR LUNCH, and there was applause.