cisco BGP aggregation bug

Thu Nov 2 04:03:12 UTC 1995

Hi there,
	We've been tracking a problem with our cisco's screwing up
aggregation of net38 at our borders.  Cisco finally issued a bug ID
for the problem today (CSCDI-43165).  We had been holding off announcing
the problem to the world until cisco confirmed where the problem is
though we have explained the situation to many that have complained
directly to us about the various side-effects of the bug and its
workarounds.
	The symptom is that our statically aggregated announcement of
net38 ("aggregate-address 38.0.0.0 255.0.0.0 summary-only") which we
had thought had been working for many months apparently had been
flapping, as seen by our neighbors (though not by us).  This was
undetected until Sprint enabled the route-dampener and discovered
that net38 was being deleted from their routing tables due to flaps.
It should be impossible, with the static aggregation, to have this
flap, we thought.  Not so, apparently.  Cisco suggested many
variations of work-arounds, some of which were not completely
successful (the most notable of which resulted in announcement of
many net38 subnets, as pointed out last week on the nanog list).
Our current config, which seems to be working fine, forces us to allow
auto summary, resulting in us withdrawing from the net39 experiment
until a real bug fix is generated by cisco.
	Our environment includes many interior routing protocols and
there may be lots of inter-related issues coming together here.  We
have not isolated the issues sufficiently to tell you what to look
for, but perhaps cisco can comment here on what might tickle this.
	At any rate, we believe the current situation is stable as
far as our announcements go.  We hope to be able to rejoin the net39
effort again soon...


							-Mitch