Re: Topological Motivation for draft-ohta-root-servers-01.txt?

[marka@isc.org]

> Mark;
> 
> > 	Geoff,
> > 		no one is try to usurp anything.  The point of the draft
> > 	is how do we provide *many* root (or tld) name servers on a single
> > 	advertised IP address and manage them effectively.  Distribute route
> > 	information, handle fault reporting. etc.
> 
> Thanks.
> 
> But,
> 
> > 	Ohta-san's premise is that every country needs a root name server.
> 
> That is a misunderstanding of you I have had no time to correct.
> 
> I think at least every contry with developped domestic internet
> needs several root name servers.
> 
> If I said every ISP needed it, there was a possibility that some
> conter argue that many ISPs can share a root server.
> 
> But, I was too optimistic, as you even say that many countries can
> share it, which will ultimately lead to the silly conclusion that the
> current number of root servers is large enough.
> 
> No, you can't convince some people who think they are receiving
> good root name service are actually receiving poor service.
> 
> So, we should not involved in the argument on how many servers
> we need and just specify a way to allow arbitrary many root name
> servers.
> 
> > 	My premise is that every body / country needs good root name service.
> 
> Geoff made a good point that several ISPs in Australia needs
> several (or, at least one?) root servers, which is different
> from your definition of "good root name service" that a
> single server is fine.
> 
> BTW, the other point of you I have had no time to correct is that delay
> is not the only factor. Fault tolerance, for example, is another
> important factor and I may be missing other factors. The best
> strategy is just specify the way to have a lot of root servers
> with underestimated reasons on why we need many without specifying
> when to have them.
> 
> 							Masataka Ohta
> 
	Ohta-san,
		I think we both agree, if we ignored borders, about
	what constitutes good root name service.  Redundant servers,
	relatively close RTT wise, sited so that single failures
	are unlikely to make the all unreachable.  The servers
	and links should also not be overloaded, this reflects back
	into RTT and packet loss.

	Just saying that every country needs a root nameserver does
	nor guarantee that people will get good root service.

	We, as a working group, should be able to define *good
	service* in measurable terms, as seen from the client.
	This sort of definition is much better that how we have
	defined root servers in the past.  It is a machine that
	can handle so many queries / second, has connections of at
	least dimmension X to at least Y places.

	Mark
--
Mark Andrews, Internet Software Consortium
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742                 INTERNET: marka@isc.org