[c-nsp] BGP causing dropped packets

[Tim Devries]

This is what I first thought of, of course, it's interesting because the
fiber circuit is provided by a third party (they use 3550's IIRC).  I have
the duplex hard-coded and no collisions etc. on the int stats on either side
(5505 is sh counters).  Also the third party has confirmed they are hard
coded to 10 full on both sides.

The interesting thing with this strange issue is that if I shutdown the eBGP
peer and ping across the 'directly connected' interface, I drop no packets,
whereas I drop bout 20% if I turn up the BGP link.  The directly connected
route is always a metric of 0 so it makes no sense. Perhaps it is an issue
with the FIB, or an IOS BGP incompatibility error?  I've upgraded to the
latest software versions with no change. Hell it's a mystery to me, maybe
one a TAC case or this list can solve...

Tim
-----Original Message-----
From: Pete Templin [mailto:petelists at templin.org] 
Sent: Friday, October 14, 2005 5:39 PM
To: Tim Devries
Cc: cisco-nsp at puck.nether.net
Subject: Re: [c-nsp] BGP causing dropped packets

Tim Devries wrote:

> The customer setup also involves another 2621 connected to another 5505,
but
> there is no issue there so I will leave the diagram as above for brevity.
> My problem is that when I shut the eBGP session down on either end, I am
> able to ping the directly attached router without any packet loss.  As
soon
> as the BGP session is established massive packet loss occurs on that link.
> I try and ping the directly connected upstream neighbor and I see packets
> being dropped all over the place.  Not so when I turn off BGP on either
side
> though.

Duplex mismatch perhaps?  Mismatch isn't triggered until BGP wants to 
route more than 200-500kbps over the link?

What do you see in "sh int" on both sides?

pt