On Jan 3, 2009, at 10:48 PM, Tony Li wrote:
Again, it's not BGP, it's the number of prefixes. Any other
protocol would
have to deal with the implications of a DFZ with a full set of
prefixes as
well. Thus, it's the architecture that must change. And, once
again, if
you don't believe that it needs to be fixed, I refer you to the
problem
statement.
Yep, and again, not strictly the number of prefixes,
but the number of paths, especially when one focuses on
RIB size and churn and implications on FIB updates and
an eye on some future security - versus sole focus on
"DFZ FIB size".
Many networks today have FIB sizes < 500k, but BGP
table sizes on the order of 5M or more. As the denseness
of inter-domain interconnection continues to increases,
the number of paths increases (more steeply than the
DFZ table size). A multi-homed AS with a PI prefix
might mean 100 *unique* paths or more in a network only
3 AS hops away. Each change in reachability for that
prefix means all those paths have to be withdraw.
Ohh, and in an ideal world, those 100 viable paths would
have to be authorized in some manner. Today this would
be done with static prefix lists at each interconnect,
although the current number of paths and absence of a
formally verifiable source for who's authorized (e.g.,
SIDR) to advertise (much less "transit" a given prefix)
what prefixes means there is essentially no inter-domain
route advertisement security in today's Internet.
More path attributes also means less efficiency in things
like update packing that allow some scaling of today's
routing system.
It's not just DFZ size that's the concern here.
-danny
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg
