[Isis-wg] operational problem might be solved by somehow extending current IS-IS

[Mikael Abrahamsson <[e-mail]>]

Hi.

I'd like to describe two operational practices and how they cause problems 
with our current IS-IS implementation.

1. When we are about to do any major work on a router (for instance now 
lately to patch the BGP bug in IOS XR), we currently set "isis overload" 
which causes most traffic to be removed from the router as all the other 
network considers it to be unavailable for transiting traffic.

2. When we want to do maintenance on a link (or we're having bit errors), 
we before used to shut down ISIS on the interface in question ("isis 
protocol shut"), but since the link might not be totally unavailable, 
we've instead started to raise the ISIS cost/metric on the link (at both 
ends) to remove traffic from it, but still have it available in case there 
is no other path (perhaps due to another failure in the network). This is 
operationally hard since we have to login to both routers which use this 
link, and do the changes on both. This is especially of interest in 
non-p-t-p links/networks, as it's hard to increase cost at "both" ends, 
because there are more than 2 nodes involved.

Basically, what I'd like to see is the way to bidirectionally increase 
cost on a link by just changing the configuration at one end. This would 
ease the number of configuration changes needed in case 2, and I would 
also request my vendor to implement a new command that might be "isis cost 
max" (or "cost avoid") that would actually not replace the current cost 
value but just raise it to a maximum value, but only needing the change 
done at one end.

To solve case 1, instead of using "overload" I'd like to use "max-metric" 
(as done in OSPF) but this would be done to all the links connecting the 
router. This would mean that in case of a BGP failure / process restart, 
ISIS/LDP signalled LSPs wouldn't have to be torn down / re-routed (torn 
down in case of a single homed router behind the router having 
maintenance). If there was a broadcast media, it would make the traffic go 
away from the router if there was any other path.

I don't really know that much about the internals in IS-IS so I don't know 
exactly how much of an issue it would be to solve the issues I've 
described above, but I'd just like to state that this is an area of IS-IS 
that would be great to have improved.

Thanks.

-- 
Mikael Abrahamsson    email: [e-mail]