Hi Francis and Ohta-san. ----- Original Message ----- From: Francis Dupont <Francis.Dupont@xxxxxxxxxxxxxxxx> Date: Wednesday, November 10, 2004 2:15 am Subject: Re: Why, technically, MIP and IPv6 can't be deployed > In your previous mail you wrote: > > > Could you describe why exactly IPv6 can't run on the (layer > 2?) WLAN > > infrastructure? > > That ND extensively, without any valid reason to do so, use > multicast, which is not acknowledged at WLAN L2, means IPv6 > or its ND is unreliable over congested WLAN. If multicast > ND packet is lost by congestion, it is not retransmitted by L2. > > => Masataka san, your argument is right (I saw 40% lost rate > on multicast over IEEE 802.11b) but it applies to IPv4 (ARP > uses broadcast) too... I understand this is the case, but we can still modify the hosts to do something about it. Multicast or Broadcast are necessary if we want generic address discovery in IPv4/6 networks. Obviously we'll have problems with MAC level acknowledgements for messages received by multiple hosts. RFC 2461 (as described by a recent individual submission of mine) allow Neighbour cache entries to be created with reliable 802.11 MAC transport (unicast toDS=0,toDS=1, multicast toDS=1) where there is only one MAC recipient in the contention domain. This may be done to preemptively place neighbour cache entries into first hop routers. Since Ohta-san mentioned MIPv6 here, the router becomes the most relevant peer for neighbour discovery. The mechanism doesn't require protocol modification, and is available today (If it's practically helpful, and implemented). It may need some further documentation though. Greg Daley draft-daley-ipv6-preempt-nd-00.txt This has been implemented in <40 lines of C in the Linux Kernel. Results, while early are fairly promising. _______________________________________________ Ietf@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ietf