BGP and convergence time

Seth Mattinen sethm at
Tue May 11 20:59:09 UTC 2010

On 5/11/2010 11:35, Jay Nakamura wrote:
> So, we have two upstreams, both coming in on Ethernet.  One of our
> switch crashed and rebooted itself.  Although we have other paths to
> egress out the network, because the router's Ethernet interface didn't
> go down, our router's BGP didn't realize the neighbor was down until
> default BGP timeout was reached.  Our upstream connectivity was out
> for couple minutes.
> I am looking for ways to detect neighbor being down faster so traffic
> can be re-routed faster.  I can do BFD internally but the issue is how
> the upstream is going to detect the outage and stop routing our
> traffic to that downed link.  I have asked both of my upstreams and
> one said they don't do anything like that, second upstream I am still
> waiting on the answer.
> My question is, do other carriers do BFD or any other means to detect
> the neighbor being down faster than normal BGP will allow?  (Both
> upstreams are major telcos [AT&T and Qwest], so I think they are less
> flexible than some others.)
> Or, has anyone succeeded in getting something done with those two carriers?

In my experience this is a pretty common problem with carrier Ethernet
links where the interface is always "up" unless the directly connected
switch/mux fails. Even then, it may still keep the port up through
reboots. I like how Ethernet is cheap, but I hate how it lacks simple
things like "link is down if any segment of the L1 or L2 between
endpoints faults" that you get without silly tricks on a DSx or OC-x.
(Then again, I suppose you're paying for that capability if it's
important enough.)


More information about the NANOG mailing list