[193951] in North American Network Operators' Group
Re: google ipv6 routes via cogent
daemon@ATHENA.MIT.EDU (Patrick W. Gilmore)
Sat Mar 4 13:37:34 2017
X-Original-To: nanog@nanog.org
From: "Patrick W. Gilmore" <patrick@ianai.net>
In-Reply-To: <20170304020515.GH1029@Vurt.local>
Date: Sat, 4 Mar 2017 13:37:30 -0500
To: NANOG list <nanog@nanog.org>
Errors-To: nanog-bounces@nanog.org
On Mar 3, 2017, at 9:05 PM, Job Snijders <job@instituut.net> wrote:
> On Fri, Mar 03, 2017 at 09:42:04AM -0500, Patrick W. Gilmore wrote:
>> On Mar 3, 2017, at 7:00 AM, Nick Hilliard <nick@foobar.org> wrote:
>>> Niels Bakker wrote:
>>>> As I explained in the rest of my email that you conveniently didn't
>>>> quote, it's so that you can selectively import routes from all your
>>>> providers in situations where your router cannot handle a full =
table.
>>>=20
>>> it can also break horribly in situations where the provider is =
providing
>>> "transit" but doesn't provide full transit.
>>>=20
>>> OTOH, if you are single-homed, it is highly advisable to accept a
>>> default, the reason being that most transit providers provide bgp
>>> communities with "don't advertise to customers" semantics. So if =
you're
>>> single-homed and use a full dfz feed without default route, you will =
not
>>> have full connectivity to all the routes available from the transit
>>> provider.
>=20
> Correct.
>=20
>> If you are single-homed, there is no need for BGP at all.
>=20
> That is very strongly worded, and in plenty of cases a false =
assertion.
>=20
>> And injecting your ASN into the table is probably not terribly useful
>> to everyone else=E2=80=99s FIB.
>=20
> ASNs don't have anything to do with FIB.
>=20
>> There are, of course, corner cases. But in general, single-homed
>> people shouldn=E2=80=99t be using BGP.
>=20
> There are numerous reasons to use BGP when single-homed:
>=20
> - as preparation to multi-home in the (near) future
> - ability to quickly change providers
> - to use BGP based blackholing features
> - to save time on provisioning work (adding new prefixes becomes a
> matter of just announcing and updating IRR/RPKI).
> - loadbalanacing / loadsharing across multiple links
> - ability to use bgp communities for traffic engineering
>=20
> In other words, if you have your own IP space, I'd recommend to get =
your
> own ASN and use BGP.
First, I said specifically there are corner cases. Everything you say =
above is a corner case. The sum of everyone in need of the above is to =
the right of the decimal compared to all single homed networks. Limiting =
it to =E2=80=9Cit you have your own IP space=E2=80=9D makes the set even =
smaller.
You are also reaching here. Preparation for multi-homing in the near =
future is just multi-homing. Adding prefixes is a very occasional thing, =
and in some cases is actually not easier with BGP. (Ever worked with =
some provider=E2=80=99s IRR implementation?) Etc.
End of day, if you have your own space and only allow aggregates into =
the DFZ, even as a stub behind someone else, it doesn=E2=80=99t really =
save RIB slots compared to having the upstream announce it for you. My =
problem is making the exceptions sound normal. They are not, and we =
should not treat them as if they are. Else we will end up with people =
wanting to do it who do not understand what they are doing, polluting =
the table, etc.
I stand by my statement. Single Homed Networks Should Not Use BGP. It is =
a good general rule. There are exceptions, but the exceptions are rare =
and should be approached with caution & clue.
--=20
TTFN,
patrick
