[j-nsp] OSPF in High-Availability environments

[Raymond Cheh]

Sending ICMP unreachable packets is independent of the routing protocol.
Will enabling BFD help? It allows fast link failure detection.

http://www.juniper.net/techpubs/software/junos/junos76/swconfig76-routin
g/html/ospf-config29.html

Thanks.

Raymond

> -----Original Message-----
> From: juniper-nsp-bounces at puck.nether.net [mailto:juniper-nsp-
> bounces at puck.nether.net] On Behalf Of Jee Kay
> Sent: Monday, June 19, 2006 10:13 AM
> To: juniper-nsp at puck.nether.net
> Subject: [j-nsp] OSPF in High-Availability environments
> 
> We have a core network of 3 Juniper M7is, all of which speak OSPF to
> each other and advertise their attached subnets into OSPF.
> 
> We're trying to get this environment as fault tolerant as possible but
> are running into a few problems... specifically, 'strange' behaviour
> during node/link failures.
> 
> I've attached a basic diagram of the layout. Essentially, when the
> primary path (labelled EES0005LON) becomes unavailable (either through
> link or node failure), OSPF recovers (hello 1, dead 3) but during the
> recovery the clients are disconnected from the servers with ICMP
> Unreachable messages.
> 
> When designing and implementing this we were doing so in the belief
> that if there are multiple paths between sites, and one path becomes
> unavailable, we would have a blackhole to traffic for a maximum of the
> OSPF dead time and after that everything would reroute itself.
> Unfortunately because the first-hop router (MSPBCR001P on the left in
> the diagram) is sending ICMP Unreachable messages after a failure of
> the primary path, the failure is very visible to the clients and hence
> to our customers.
> 
> Is there some particular configuration I should be using to prevent
> OSPF/the router sending out these ICMP Unreachables (or are there any
> 'How to configure Junipers and OSPF in fault-tolerant environments'
> guides)? Or is there simply no way to have seamless fault tolerance
> with Junipers in our core?
> 
> Thanks in advance,
> Ras