On Mar 26, 2009, at 12:31 PM, Gargi Nalawade wrote:
Disclaimer : Im not taking a stance on whether it is good or bad for
an RR to advertise
a client's routes back to it. Want to make some observations on the
pros and cons that
go into the advertise or not-to-advertise decision.
If the RRs were to stop reflecting back own routes to the clients,
this would break
any peer-groups configured on the RRs. This would result in a lot
more individual
BGP updates being formatted and sent out. In the worst case 100
different sets of
updates for 100 clients. This would slow down convergence.
That's an implementation issue and practical reality
that folks have to deal with, I understand that. It's
a matter of when the pain threshold exceeds the value
of this optimization.
Interestingly, this is how Juniper implemented it
initially - I'm not sure what the current behavior
is but I know that ACCEPT_OWN would require them to
change at least _some of this, if they haven't already.
I know this because a C router was reflecting routes
back to a J router and the J router was installing, and
an interesting discussion commenced at the time in a
network where I was an operator. This is when J
implemented the poison-based-on-orig-ID on the client
side. I hope it didn't drive them to change the route
reflectors 'reflection' behavior as well.
-danny
_______________________________________________
GROW mailing list
GROW@ietf.org
https://www.ietf.org/mailman/listinfo/grow
