Mailing List Archive

ipv6-ops Digest, Vol 2, Issue 9
I presume that the "no decent service...by Abilene" statement
is overstated enough that it doesn't need answering? I hope
that there's *some* benefit... :|

It's true that so far Abilene's only direct commercial peerings
are on the west coast, but that's not an issue of "capability",
just current status. We're at exchange points in Chicago (Star
Light), NYC (Manlan), and have indirect access to the NAP of the
Americas in Miami through AMPATH. We've also been trying for some
time to find additional east-coast v6 commercial peers but haven't
been able yet to make anything work out (and we'll continue to
try). That doesn't preclude some of our university members in
the Eastern US from passing commercial v6 prefixes to the rest of
Abilene, though I'm not aware of any at the moment who do. Since
almost all of them have broader local infrastructures than the
few Abilene POPs, they could bring in peerings that we can't reach.
So yes, because at the moment Abilene's commercial v6 peerings
are all on the west coast, peers who use Abilene to get from the
east coast or Europe to European commercial networks have to cross
the US. (that's not true for R&E prefixes, which connect and are
transited freely at all four borders.)

To be practical, how can this be improved? obviously additional
east-coast v6 peerings for Abilene would help immensely (anyone at
32 AoA want to talk?) Perhaps, in the absence of that, European
R&E nets might want to choose better paths to some US commercial
v6 customers than via Abilene, but as you point out below that's
chiefly a dis- advantage for east-coast users, not west-coast
ones, and many foreign networks probably can't distinguish a part
of a distant network very well from another part. We'd be happy
to improve it. Thanks for your suggestions.
--
Brent Sweeny
Indiana University Information Technology/Abilene Network Operations
PGP fingerprint = 0F 6B 7E 1D 3A AD F3 01 63 1E 2B B3 1E B1 FA 7F

> Date: Tue, 24 May 2005 22:23:40 -0400
> From: James <james@towardex.com>
> Subject: Re: Canarie / 2001:410::/32 (Was: Filters) (fwd)
> Cc: ipv6-ops@lists.cluenet.de
>
> On Tue, May 24, 2005 at 08:25:05PM +0200, Bernhard Schmidt wrote:
>
> [ snip ]
>
> >
> > I can't currently see any decent service provided by Abilene. GEANT
> > being the european REN isn't that much better.
>
> Now that we are talking about RENs :) One of the GEANT routing policies[1]
> I've heard from several sources is that they will take commercial transit
> (GBLX) for European v6 destinations, but yet, take Abilene for all non-EU
> routes, i.e. US destinations.
>
> While I agree that this may be a sane choice of policy, I would hope that
> people understand that Abilene currently has capability to peer with
> commercial/commodity v6 players in the west coast US only (PAIX-PA, and
> possibly LAIIX/PacWave). Being a US network myself, this gives me quite
> *bad* routing to European NREN destinations for our end-users who reside
> in the east coast of United States, while it is fine for west coast users,
> as return packets from European RENs go all the way to west coast via
> Abilene, then come back to east.
>
> What is even more interesting is that this "take Abilene for US routes"
> "policy" is not just GEANT only, it also seems to be exercised by regional
> EU NRENs as well (i.e. SURFnet), even though such regional networks have
> direct transit from C&W or other similar commodity transit.
>
>
> [1]: Accuracy of this statement is not guaranteed and I may be wrong.
> So please feel free to correct as needed.
>
> -J
>
> --
> James Jun
> Infrastructure and Technology Services
> TowardEX Technologies
> Office +1-617-459-4051 x179 | Mobile +1-978-394-2867
> james@towardex.com | www.towardex.com
ipv6-ops Digest, Vol 2, Issue 9 [ In reply to ]
Hi Brent,

> I presume that the "no decent service...by Abilene" statement
> is overstated enough that it doesn't need answering? I hope
> that there's *some* benefit... :|

Well, I agree that it was a little mit overstated, but not completely.
Since there currently is no IPv6 killer application in sight an
important task in the IPv6 transaction would be making services and
clients dualstacked and see what happens. Unfortunately this is pure
poison for the transaction process when the client suddenly speaks IPv6
and the connectivity is that bad that the user cries "it's so slow, stop
this damn IPv6 I can't work anymore".

Now for the cause of this bad connectivity. I agree that connectivity
_is_ decent (more than that) for NREN members and customers of the few
commercial peers I can make out. But, for the whole remaining networks,
it just plain sucks. Have a look at GRH at Abilene's prefix, a _large_
cloud of networks see something starting like this as best path to them

6939 4716 11537

There we have America-America going through some Asian prefix. 6939 on
itself, which seems to be the only way out of 4716, is deprefed at many
operators due to needless fulltable-swaps and bad long-range tunnels, so
not really good connectivity there, too. Some other paths:

29670 12732 20646 1752 5511 2500 4725 11537
3292 6175 4555 6939 4716 11537
12871 24587 6453 10566 17715 4725 11537

I think one can let those paths speak for themselves.

The other direction isn't much better unfortunately, since Abilene again
prefers an Asian ASN (prolly NREN) to go to European networks. Just a
few examples pulled from the GRH output:

2001:650::/32 11537 17579 9270 7660 2500 1273
2001:668::/32 11537 17579 9270 7660 2500 2497 3257
2001:6c8::/32 11537 17579 1237 17832 9270 7660 2500 4691 2914 3292
2001:788::/32 11537 17579 9270 2200 20965 1299 1759

> It's true that so far Abilene's only direct commercial peerings
> are on the west coast, but that's not an issue of "capability",

Peering isn't the only solution. You can peer with dozens of networks,
as long as you don't have _upstream_ some commercial networks will see
you through the peering, but the great remains will see you through some
routeswaps on another continent.

GEANT did the only solution for this kind of situation, they got
commercial _transit_ from Global Crossing and Telia, so as long as they
don't send traffic to European prefixes to Abilene by mistake their
connectivity is actually quite good.

Current state from a few boxes I maintain:
64 bytes from ping-wash.abilene.ucaid.edu: icmp_seq=1 ttl=34 time=421 ms
64 bytes from ping-wash.abilene.ucaid.edu: icmp_seq=1 ttl=34 time=393 ms
64 bytes from ping-wash.abilene.ucaid.edu: icmp_seq=1 ttl=38 time=432 ms

all of them have decent commercial transit, but even they only get the
prefix through Asia.

We can move this discussion to private mail if you want, but I do
support my claim that IPv6 support at NRENs is improvable, to say the least.

Bernhard
ipv6-ops Digest, Vol 2, Issue 9 [ In reply to ]
On Wed, May 25, 2005 at 10:04:43PM -0500, Brent Sweeny wrote:
> To be practical, how can this be improved?

- get one/two decent upstream(s) with good coverage of US and
especially EU. When that is in place, drop full table from
APAN/KREONET2 as you cannot differentiate between commercial and
R&E routes from them.

- make sure that all commercial routes are tagged ingress so that
they'll never leak to any other commercial peers or transits.
Don't send full table to commercial peers and upstreams, but only
the set of Abilene member routes.

- try to show up on NYIIX, Equinix Ashburn, LAIIX and NOTA (you said
you're there "via AMPATH"... is that L2 or thru some L3 arrangement?)

Those three things (even only the first two!) would improve the
situation _big_time_.

> We'd be happy to improve it. Thanks for your suggestions.

Great to hear from you. I'm positive that this all gets sorted out.
:-)

Thanks for taking the time to engange in ipv6-ops.


Best regards,
Daniel

--
CLUE-RIPE -- Jabber: dr@cluenet.de -- dr@IRCnet -- PGP: 0xA85C8AA0