[c-nsp] Best practice for lowest convergence time when rebooting router

[Pete Templin]

Jay Nakamura wrote:
> Quick question,
> 
> Does anyone do anything special before rebooting/shutting down a core or
> border router so routes on other routers converge faster around the rebooted
> router?  I was thinking if it would be better if I heavily BGP prepend to
> upstream first so bordering peer router will still route to me until the
> better route from the other upstream propagated to that peer.  What about
> OSPF inside our network?  Putting the interfaces in passive mode first?

On our "edge" routers (nodes that handle upstream connectivity), we have 
pre-made route maps for each provider in multiple forms: normal, "high 
med", and maintenance.  We switch out the route maps ahead of provider 
maintenance or internal operations, and restore when we feel comfortable 
that the work is complete and stable.  Do note that for most transit 
connections, prepends aren't sufficient to push inbound traffic off the 
link; you'll normally need to signal peer-level or peer-backup local 
preference with communities.

On our "core" routers (nodes that interconnect POPs as well as edge and 
distribution routers within the POPs), we do "max-metric router-lsa" 
within OSPF to push traffic off the affected node before 
maintenance/reboot.  Once complete, we return to "max-metric router-lsa 
on-startup 900", so that the router won't take transitory traffic 
(unless necessary) for 15 minutes after a reboot, providing an 
opportunity to assess conditions.  For reloads, we schedule the reload, 
then change the config to the simple form of "max-metric router-lsa" but 
don't re-save the config.  This way, the router returns to service 
unattended.

pt