retitling this because I really, really, really, would like more people using hncpd and providing feedback on that, rather than arguing over specification documents.
On Fri, Feb 20, 2015 at 2:14 PM, Markus Stenberg <markus.stenb...@iki.fi> wrote: >> On 20.2.2015, at 22.01, Dave Taht <dave.t...@gmail.com> wrote: >> On Fri, Feb 20, 2015 at 11:34 AM, Steven Barth <cy...@openwrt.org> wrote: >>> >>> Am 19. Februar 2015 20:05:56 MEZ, schrieb Ted Lemon <mel...@fugue.com>: >>>> Hm, I will have to try it out. Is it in a distribution? >>> >>> ohybridproxy in openwrt. It's mainly useful with hnetd (hncp) though. >>> >>> Manual configuration without hncp is a bit awkward since you need to name >>> each link manually and on every router configure the resolver (e.g. >>> dnsmasq). I guess we might want to provide a little example for 2 links at >>> some point. >> I would like to deploy hncp in my upcoming make-wifi-fast testbed. >> However the biggest headache is ensuring that all the routers >> inbetween have hncp burned into them, and are only acting as relays as >> I generally only obtain a few (/60) real IPv6 prefixes per GW and they >> only need to be on the APs. >> >> GW1 - routerA - routerB - routerC - routerD - AP1 >> | >> GW2 - routerE - routerF - routerG - routerH - AP2 >> | >> AP3 >> >> GW3 ... >> >> (the actual topology of the testbed is way more complex than this >> (covering ethernet, wifi, and moca) and I am not going into it here) >> >> 1) is there a way to configure hncpd as purely a relay, and NOT do any >> address assignment at all on routers B,C,D,F,G,H? > > In theory (=spec), but not in current implementation (I _think_). Yes, the spec seemed right, but after playing with the code, got stuck, so I thought I would ask. If you can point me at any specific area of the code to try to make relaying work, I can dig into it; I am about a month away from deployment. (and have other things I would rather do than improve hnetd, like make-wifi-fast) I think having hncp relaying support universally is needed, and for devices that are homenet participating but not necessarily full blown devices - should be configured as a "relay only" by default. the local multicast proxying problem is partially why pim-sm and related protocols did not deploy to a huge extent, IGMP notoriously buggy, etc, etc. See for example, the sad, underused state of pimd. > You do not really need non-linklocal addresses for HNCP, or routing protocol, > so as long as there are no hosts on the link.. That to me, was a truly major point of both hncp and routing protocol. I have seen plenty of people gripe elsewhere about assigning *any* public ipv6 addresses to a home router, preferring that dnsmasq´s dhcpv6 server work without one, for example. >(traceroute is a bitch though given just linklocals) I don´t care about traceroute. > > As workaround in current implementation, if you set it to assign e.g. /80 per > link used for your intermediate links, you would have almost all of address > space left; however, in your case, I am not sure you can even afford to split > one /64 for that purpose. Nope, 22+ APs, most with 2 or more subnets, and a couple /60s to play with. Have no idea with the topology above how IPv6 prefix exaustion will be handled either. But - more importantly - I am not interested in exposing intermediate routers to the wild and wooly internet with public ipv6s. Got enough firewalling problems as it is. I realize that my use cases are a bit different than standard homenet´s, but I am very interested in exercising the protocol to a fuller extent than it has before. >> 2) have you tested that it is indeed possible to get the separate ipv6 >> prefixes from GW1,GW2,GW3 to AP1,AP2, AP3? > > Yes. > >> 3) Can ULA and "Real" address assignment be distinguished along the >> way? I have no problem assigning ULAs to the routers, but dont want to >> use up real addresses on them. > > In theory (=spec), not in current implementation (ULA is actively discouraged > in current impl, I cannot remember if there was option to override). What I had done with ahcpd was: it calls a script to configure every received address, filter out anything but ulas, and just configure those. Easy, couple lines of code. > >> 4) What happens when someone pulls the plug on GW1, it reboots, and >> gets a new Ipv6 subnet (I have seen comcast do this to me >> every time that happens with the code I have in place now - no >> retraction, and I go through hell manually eliminating every former >> prefix from the network. Yes. I have upses. And cerowrt, at least, >> stays up for 90+ days without a problem. But it happens and sucks when >> it does) > > Renumbering should just work as usual. I.e. rest of nodes should learn of the > new prefix, and old one should disappear. I hope so. Arbitrary renumber is really the biggest PITA I presently have with comcast. Well, with the exception of naming - ipv6 mapping - which I have just totally given up on at present. Try to get a printer to reliably work with a dynamic public ipv6 address... with mdns... it is less painful to merely shoot nails into your head. > -Markus -- Dave Täht thttp://www.bufferbloat.net/projects/bloat/wiki/Upcoming_Talks _______________________________________________ homenet mailing list homenet@ietf.org https://www.ietf.org/mailman/listinfo/homenet