Seeing Teredo traffic from Vista?

[Bernhard Schmidt]

On Fri, Feb 16, 2007 at 06:49:17PM +0100, Daniel Roesen wrote:

> On Fri, Feb 16, 2007 at 04:22:15PM +0000, Carlos Friacas wrote:
> > GEANT2 prefers Abilene (11537) routes..... "educational" routes have 
> > bigger preference over commercial ones.
> > In this case the result is not brilliant...
> The result is seldom brilliant. It might make sense to prefer edu
> transits to edu routes, but not to prefer asia edu transits to
> european/us commerical routes.

I hope you do have your barf bag ready, because almost all european
NRENs take paths to quite a number of prefixes (around 15 /32s, most
probably a high number of more-specifics as well) like this

2001:440::/32   680 20965 24489 23911 23910 4538 7660 2500 2914 6175 i

in words: DFN (german NREN) - GEANT2 (European REN) - TEIN2 - CNGI-BJIX
- CNGI-CERNET2 - CERNET - APAN-JP - WIDE - NTT/Verio - Sprint

while AS2500 is known as a source of bad paths for ages now (usually
leaking paths learned from 2914, 7660 or 4725 through 2497 to 3257 or
through 18084 to 3549), this direction is quite new. It appears that
those five consecutive Asian NRENs don't get their filtering along,
GEANT is accepting that and (worst!) distributing that as a customer
route to its upstream Telia (and Telia is accepting it, strike three).
Yes, GEANT has been informed and refused to change it.

Do a "sh bgp ipv6 unicast regexp _2500_" on your local table, it is not
pretty.

> But that's no news. Unfortunately those asian "full table providers"
> (last time I looked and discussed the issues with Abilene) don't tag
> their routes properly so that other edu networks can differentiate
> wether a destination route is via commercial transit in the end or
> not.

Well, if it is a known broken behaviour, both GEANT and Abilene should
resort to doing strict prefix-based filtering on their links to those
"offenders" until the issues are fixed.

If I choose that path, I have whooping 800ms RTT when I reach Sprint.
How are we supposed to ever do more than traceroutes with that?

> Overall it would be good if the edus would use their local commercial
> transit for commercial destinations(!), and their edu peers for edu
> destinations(!).

Ack.

Regards,
Bernhard