[nsp] 7206/NPE-G1, 12.3.8T and BGP slow convergence

[Łukasz Bromirski]

Hi,

I have a problem with 7206VXR/NPE-G1 router, running 12.3.8T with
Plus IPsec/3DES only. It's a border router, with one neighbor only.

My provider put another typical neighbor entries on his side, I've
put one on mine, and this is where problems start.

We're using `ip tcp path-mtu-discovery', and the session get's MTU
of 1440 bytes. It should be 1460 IMHO, as we don't have any other
routers in way, but there is some mysterious multiplexer, that is
doing `transparent' Ethernet service from two E1 links.

And, is my observation correct, that most of the BGP packets come
not in sequence?

----
Datagrams (max data segment is 1440 bytes):
Rcvd: 215 (out of order: 180), with data: 89, total data bytes: 16732
----

Providers bgp neigh shows many standing updates in the out queue, and
it looks like a bandwidth/queueing problem on their side, but as
pings with DF set and lenght of 1500 bytes go there and back, what
can be the cause of such caveat?

I've tried to add selective acks, suspecting some weird TCP
resynch problem, but it doesn't help.

The physical bandwidth is not a problem - it's a 4Mbit/s link and
can actually transport that amount of data, and we're using for
now about 200-250kbit/s.

Another thing is, routers will exchange ~1000-1200 prefixes in
25-30 minutes, and then the session drops (and is established
again):

%BGP-5-ADJCHANGE: neighbor A.B.C.D Down BGP Notification sent
%BGP-3-NOTIFICATION: sent to neighbor A.B.C.D 4/0
                      (hold time expired) 0 bytes

Does someone run 12.3.8T on NPE-G1 and can comment on BGP behaviour
in his/her network? Or someone fought similar problem?

-- 
this space was intentionally left blank    |          Łukasz Bromirski
you can insert your favourite quote here   |      lukasz:bromirski,net