To:
"Tim Chown" <tjc@ecs.soton.ac.uk>, <dnsop@cafax.se>
From:
"BELOEIL Luc FTRD/DMI/CAE" <luc.beloeil@francetelecom.com>
Date:
Thu, 17 Jul 2003 11:29:19 +0200
content-class:
urn:content-classes:message
Sender:
owner-dnsop@cafax.se
Thread-Index:
AcNLvqT1Cc3nQwuSR1iqSLnvmW0bSgAgcWcQ
Thread-Topic:
TR : Stepping back on the DNS discovery discussion
Subject:
RE : TR : Stepping back on the DNS discovery discussion
Hi Tim, Thanks you to react., My coments below > -----Message d'origine----- > De : Tim Chown [mailto:tjc@ecs.soton.ac.uk] > Envoyé : mercredi 16 juillet 2003 19:04 > À : dnsop@cafax.se > Objet : Re: TR : Stepping back on the DNS discovery discussion > > > On Wed, Jul 16, 2003 at 02:31:14PM +0200, BELOEIL Luc > FTRD/DMI/CAE wrote: > > > > As DNS is a critical service between "applications world" and "IP > > layer world", we might be careful on consequences. From application > > point of view, we might ask in the future to extende that > solution to > > other type of services (NTP, ...). In that way, RA-based solution > > might not be the good solution, because I do think that RA > must not be > > extended to any sort of service discovery. Any idea about that? > > So where do you draw the boundary between "network device > autoconfiguration" > and "service discovery". For IPv4, you (manually or via > DHCP) configure > IPv4 address, netmask, gateway and DNS resolver(s) as the > "basic" four > components to be able to get up and running. In IPv6, when > using stateless > autoconfiguration, the RA mechanism takes care of the first > three (because > the stateless mechanism implies the /64 "netmask"), but the > DNS resolver > address(es) cannot be learnt (although I believe WinXP implements the > well-known site local discovery method). > Yes indeed > So an interesting question is what we mean by "stateless > autoconfiguration". > I think this term is causing some of the religion in the > discussion :) It > seems that probably mean "autoconfiguration without servers > (i.e. DHCPv6)", since there would have to be state > (pre-configuration) in the router (the > prefix configuration). The RA camp seems to want the > routers to be able to > issue DNS resolver information, not a separate full-blown > DHCPv6 server. > Hum as far as I understand DHCP-lite solution,DHCP-lite is also a stateless solution. > But it's not that clear cut as, for example, my home DSL > router includes a DHCP server (as well my DNS server and > print server and other stuff). The functionality is combined, > all in a very cheap (<100 Euro) commodity box. > How much a solution is , is only one side of the issue, not all -it depends so much of the market ;+). I do not discuss that. > So are we really concerned as to whether we need a separate > DHCPv6 server on our network (and the overhead of that in > ad-hoc networks), or that we would > like a router to be able to provide DNS resolver information > on link (such that > we can plumb networks without needing a separate DHCPv6 box)? > If the latter, > then does it matter if the router does that via RA piggybacks > or DHCPv6 Lite? What is the real difference in implementation cost? > Good question, I don't still know. > Of course the mode of operation is also an interesting issue, > but it's possible for DHCPv6(Lite) to have an unsolicited > response mode added, if that is felt to be important. > That solution must be investigate also. > Something has to configure the prefix information to the > router, likewise the DNS information has to be configured. I > don't think we've discusse that much, but distributing the > DNS resolver information to multiple routers rather than one > (or a small number of) DHCPv6 servers may require more > thought. But in my home network, it'll all be in my DSL router :) > Some questions must still be investigated so that I could have a better opinion. - Can DHCP-lite work on the same link were there is a DHCP server, or a DHCP-relay? - is DHCLP-lite with multicast extension is still DHCP? ;+) - is RA-base solution actually much easier to implement? - How RA-based wil be implement, within the kernel, as a separate daemon? ... > Tim Luc > #------------------------------------------------------------- > --------- > # To unsubscribe, send a message to <dnsop-request@cafax.se>. > #---------------------------------------------------------------------- # To unsubscribe, send a message to <dnsop-request@cafax.se>.