Excerpts from Pekka Nikander on Mon, Jan 19, 2009 08:23:19AM +0100: > One of the repeated arguments against host-based solution is that it > increases signalling load. Of course, that is true for a pure host- > based solution where each host individually maintains the mapping > from EIDs to RLOCs. So, from a signalling load point of view, such > a solution is not very good if the mapping granularity is at the > site level. > > There is a relatively easy solution to this, which brings the > inter-site signalling load to the level of tunnel-router-like > maps-and-encaps solution. The basic idea is that the hosts > *delegate* the signalling to a tunnel-router-like device at the site > border. (At the same time, the hosts can still signal individually > through another interface; consider e.g. a laptop which is attached > with a wire to a corporate site but at the same time has a wireless > interface using an open WLAN.) > > So, when both a host and a site support a (multi-homing/mobility) > solution, the following can be done: > > 1. The host detects that the site supports the solution, e.g. > through DHCP/ND. > 2. The host delegates any mobility/multihoming/tunnelling signalling > to the TR at the site. > 3. The site border TR/whatever takes care of signalling on the > host's behalf.
Pekka, I don't think this is a win. As an endpoint, you are interested in reachability of specific endpoints. You don't care if a whole prefix is reachable, if the particular endpoint isn't reachable within that prefix. So if you delegate liveness monitoring to an agent, you are asking it to monitor specific locators. First of all the number of locators that need to be monitored does not decrease. The problem is a lack of parallelism between the different sides of the transaction. Intermediaries like ITRs and ETRs can monitor the health of interfaces to whole prefixes, since that is what they are responsible for. Endpoints are responsible for communication with other endpoints, not just prefixes, so that is the information they need. Second, as Fred Baker pointed out in many cases exchanges are short and you don't need probing or monitoring, and in some cases there may be different "grades" of continuity required which may allow "relaxed" monitoring ... but a liveness agent won't be able to keep track of that nearly as well as endpoints, so it will over-probe unnecessarily. Does that make sense? Do I understand correctly? > When considering HIP, the theory is discussed in length in Jukka > Ylitalo's Ph.D. http://lib.tkk.fi/Diss/2008/isbn9789512295319/ > > For a practical example, see Section 4 of draft-melen-hip-mr-01.txt > > For measurement results, see Jukka Ylitalo, Jan Melén, Patrik > Salmela, and Henrik Petander, "An Experimental Evaluation of a HIP > based Network Mobility Scheme", in Proc. of the 6th International > Conference on Wired/Wireless Internet Communications (WWIC 2008), > pp. 139-151, I'll take a look. Scott _______________________________________________ rrg mailing list [email protected] http://www.irtf.org/mailman/listinfo/rrg
