[83833] in North American Network Operators' Group
Re: {f,i,k}.root-servers.net anycast instances deployed in India
daemon@ATHENA.MIT.EDU (Suresh Ramasubramanian)
Fri Aug 26 21:22:01 2005
Date: Sat, 27 Aug 2005 06:51:34 +0530
From: Suresh Ramasubramanian <ops.lists@gmail.com>
To: Steve Gibbard <scg@gibbard.org>
Cc: NANOG <nanog@merit.edu>
In-Reply-To: <20050826120643.S23822@sprockets.gibbard.org>
Errors-To: owner-nanog@merit.edu
On 27/08/05, Steve Gibbard <scg@gibbard.org> wrote:
>=20
> If we look at the Asia-Pacific region (for these purposes everything east
> of the UAE and West of the Americas), and then exclude Japan, Korea, and
> Singapore, countries that are undisputably part of the Internet core, wha=
t
> we've got are a bunch of F and I Roots, with a K Root in Brisbane and now
> a K Root somewhere in India. Having root servers that are part of three
> different anycast clouds would make India somewhat special within its
> region.
>=20
The crying shame of it all is that most ISPs, for various reasons
[below] don't advertise all their routes at nixi - and so exchange
piddly little amounts of traffic where they could exchange LOTS more.=20
So, you'd find a whole lot of Indian traceroutes, even between two
local ISPs, go out through Singapore (or possibly Reach / NTT now, in
some cases), and/or PAIX
Having three anycast instances in a country is no damned use when the
nearest roots, network wise, are elsewhere. If traffic stats from
those things are available, and are studied, I am reasonably sure we'd
get some interesting results.
But for now, having X number of anycast roots in the country is only
scoring brownie points in the i-governance debate.
--srs
Reasons include -
* don't have good people with bgp clue, only "senior network admins"
who ask Philip Smith what a route map is, in an advanced bgp tutorial
at a recent SANOG...
* don't trust each other too much at all, and fear that peering means
that people can rip them off by using those links for transit as well
* have a network that's a mess of botched mpls and other
implementations, all held together by a bunch of static routes
upstream
* or in some cases have, besides their usual IP space, a huge lot of
deaggregated IP blocks that are a legacy from when they had a whole
lot of leased lines purchased from the then incumbent + sole upstream
VSNL ..
