[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]


To: Harald@Alvestrand.no (Harald Alvestrand)
Cc: hardie@equinix.com, kre@munnari.OZ.AU (Robert Elz), dnsop@cafax.se, aroot@ops.ietf.org
From: hardie@equinix.com
Date: Fri, 3 Nov 2000 14:18:33 -0800 (PST)
In-Reply-To: <4.3.2.7.2.20001103223159.06555a40@127.0.0.1> from "Harald Alvestrand" at Nov 03, 2000 10:45:27 PM
Reply-to: hardie@equinix.com
Sender: owner-dnsop@cafax.se
Subject: Re: Single global announcement (Re: Anycast root metrics and

> 
> If read as a physically single point, the single global announcement idea 
> kills any hope of the anycast scheme benefiting users in smaller ISPs 
> outside the US.

The single global announcement doesn't have to mean a single physical
presence; it could be combined with the anycast hack described in my
draft to distribute the servers around the global announcer's
topology.  Frankly, though, I would not combine the two.  Given that
there are multiple root servers, I would rather suggest doing one with
one root server and the other with a different root server.  That
strategy could mean, however, that the ISPs not running their own copy
would be dependent for good performance on a connection to some entity
large enough to have a network presence pretty much everywhere.  That
tends to mean fee-based traffic exchange.

> 
> I still think an anycast AS number needs to be centrally owned and 
> administered, with monitoring of the global routing tables to detect bogon 
> announcements, just like a real, normally internally-connected AS would behave.
> 
> We might agree.
> (and - I think the Right Thing is a Root Server Consortium operating under 
> ICANN contract, with responsibility for ONE of the addresses, at least at 
> first, with the rest of the root server set being managed exactly the way 
> it is today. but we are getting VERY far into politics rather than technology).


A centrally administered anycast AS has a much better chance of
imitating the working of an internally-connected AS than Masataka's
current proposal or the compromise I suggested.  Depending on the
implementation, it might not look very different from those multihomed
enterprises that announce their address space via BGP and that do
internal connections via VPNs.

Getting that to work in a root server consortium though does push us into
the politics of how does get approval to join, blah, blah, blah.

				regards,	
					Ted Hardie

Home | Date list | Subject list