nanog mailing list archives

Re: BGP 011: multiple sessions with upstreams


From: "Christopher L. Morrow" <christopher.morrow () mci com>
Date: Sat, 01 Jan 2005 05:07:55 +0000 (GMT)



note, me == chemical engineer, someone else here is a network engineer :)

On Fri, 31 Dec 2004, Edward B. Dreger wrote:

      Would you please provide in detail your reasoning for needing
      the two BGP sessions and also why you would not need two
      sessions with [other upstreams].

I'm trying to persuade them that two provider/customer BGP sessions is a
good thing, and NEXT_HOP set to the HSRP-managed IP address would allow

I'd start with:
1) hsrp allows the ethernet to failover 'gracefully'
2) there are no provisions for non-stop-routing in your above description
3) nothing is there to maintain the protocol state between your 2 devices
and their 1 device
4) how is traffic from YOU to ISP supposed to flow given the dependency on
the BGP session which will surely fail when one of your 2 routers dies?

them to steer customer traffic.  Their argument is that one can "get
extra bandwidth" with two BGP sessions, and that HSRP prevents that.

you can't get more than the bandwidth on the link... they are metering
that so why do they care? 1mb to 1 destination or .5 to 2.. who cares?
They will judge by the stats on their interface regardless of the number
of destinations on your side of the ethernet link.

I've pointed out that equal-cost multipath BGP is far from default
behavior, and that one can "get extra bandwidth" even without two BGP
sessions.

you are not asking for equal cost multipath, you are asking for 'shadow'
(sort of) bgp...


Am I missing something?  I've attempted to stress the need for
redundancy, and highlighted that they themselves have a couple border

redundancy in the ethernet you've covered, you probably need to walk them
down the 'if my ethernet goes, my bgp goes, so my forwarding to YOU
goes... unless there is bgp elsewhere from you, hence my request for bgp
from 2 routers to your router' path.

routers which presumably have dynamic routing sessions to both edge
switches.




Current thread: