draft-ietf-ptomaine-nopeer-00

[Joe Provo]

On Wed, Nov 20, 2002 at 03:07:08PM -0500, Jeffrey Haas wrote:
> On Wed, Nov 20, 2002 at 02:31:43PM -0500, Tony Tauber wrote:
> > So, would it count if someone were already to have a community (or
> > two) that customers can use to control the redist. of their prefixes,
> > but it's not the well-known one?
> 
> The problem with those is that this is meant to be far more transitive
> than other internal communities.  I would suspect that this may
> reach 2-3 as hops away from the person who attaches the community
> while current globally well known ones, such as NO_EXPORT, etc.
> only go as far as the first AS.
 
Transitive comunities do exist in multi-AS providers' environments 
(701 customers can influence route policy at edges of 702, etc).  
Considering the number of "make me prepend at my edge", "send to 
customers only", "send to customers but not tagged as your customer",
etc communities that are deployed in provider-specific manners, end 
users are counting up as-paths, making policy configurations and 
tweaks per-provider.  Certainly the ability to smooth out the 
per-provider configurations can only help reduce complexity and 
therefore reduce the oft-cited operator-induced errors?

Does that make it "best current"? I don't know - I stayed out of 
arguments of that lingo when 1597 and 1627 begat 1918.  The only 
"current" I can think of off the top is 70x example; I know Level3
had communities that were transitive between their ASNs as well, but
they no longer run multi-ASN. 

Does this translate to people operating at a distance through peers? 
I don't know, but there are networks who do handle MEDs and 
deaggregates with peers on an as-agreed basis. Would that be considered 
prior art for operating at a distance through peers?

Cheers,

Joe

-- 
 crimson at sidehack.gweep.net * jprovo at gnu.ai.mit.edu * jzp at rsuc.gweep.net
             RSUC / GweepNet / Spunk / FnB / Usenix / SAGE