AS690 Gated/BGP4 Deployment Status

Sun Feb 13 14:35:22 UTC 1994

AS690 Gated Installation Summary - Sunday Feb. 13th 08:10EST
================================
The AS690 gated deployment this past evening was very successful.  We
are now running gated in production on ENSS205, ENSS194, ENSS160,
ENSS131, ENSS139, CNSS120, and ENSS158. 

These gated nodes are now supporting interoperability with the
rcp_routed IGP & IBGP2, external BGP2 and external BGP4 (3 peers at
Rice U. thanks to Bill Manning), and several EGP peers.  We also are
monitoring the appropriate gated MIBs.

We found some minor problems along the way, some of which will
have to be fixed before the next scheduled deployment, but we did
not see anything that would result in operational problems, or require
us to back out to rcp_routed.  

We would like to schedule the next deployment for Tuesday morning Feb
15th (05:00-08:00EST).  We would like to deploy gated on ENSS136,
ENSS145, ENSS144 during this window.

Once we successfully complete installation on these nodes, we would
like to deploy across the rest of the AS690 system.  This will require
a bit of work on the Policy Routing Database to ensure that don't have
to do any manual corrections to the gated configuration files as we go
along.

The summary of problems that were observed during the Sunday
morning gated deployment include:

1.   Gated "passive" connection option not working.  Rcp_routed does
     not actively try to establish sessions with external routers, and
     instead waits for them to establish the connection.  Gated tries
     to actively connect to all configured external peers unless the
     "passive" configuration option is used.  The passive
     configuration option in gated does not seem to be working, and we
     observed a few startup connection wars occur between ENSS131 and
     its BGP peers, and likewise with ENSS139.  The problem occurs
     when gated and its external peers try to establish connections at
     the same time.  This settles down after a couple of minutes and
     the connections stay up once established.  This will be fixed in
     gated.

2.   We observed on ENSS194, and ENSS139 that upon some
     occaisional initializations of gated, we sometimes do not get all
     of the LSP packets, and the link state database is incomplete. 
     In the case of ENSS194, it did not get all of its adjacencies
     when gated was first started.  This worked itself out on its own
     after a few minutes.  On ENSS139 we saw the same thing, only
     it did not work itself out until after we re-started gated.  This
     most likely has something to do with way rcp_routed establishes
     adjacencies.  We did not see this on the testnet, and we have
     no way easy way to debug this, but once stable, the gated
     systems seem to stay that way.  For now, we will address this
     problem by restarting gated if it gets in this state following gated
     initialization, and migrate away from rcp_routed as soon as
     possible.

3.   We found a bug in the gated dynamic reconfiguration where
     gated will crash if we try to reconfigure on the fly with gated
     systems that support multiple EGP peers.  Rcp_routed does not
     support the same level of dynamic reconfiguration as gated. 
     We were able to reproduce this problem consistently on
     ENSS139 (Houston).  This is a bug that we would like to fix
     before the next wave of gated deployment.