OSPF and EIGRP Graceful Restart (NSF) clarification

[vincehgov]

Hi there,

I'm studying for my CCIE R&S and I just need some clarrification regarding OSPF and EIGRP graceful restart.

I just wanted to confirm that when the RFC and the Cisco docs say "restart", they are talking about the router doing a full reload as in what would happen when I issue a 'reload' command.

This would mean that the hardware is running on its own to perform routing (using the hardware FIB) while the software is reloaded.  And, that the OSPF software somehow knows that it is in graceful restart mode after it gets reloaded in order to go through the rest of the graceful restart procedure.  And, that the physical interfaces don't go into 'down/down' mode when the router is restarting- meaning that the rest of the router hardware knows how to keep forwarding regardless of the fact that the software is reloading.

Am I right to assume all these things?

I have read the cisco docs and RFC but just need someone to confirm my logic here.

Vince

[Giuseppe Larosa · Accepted solution]

Hello Vincent,

first of all, we need to distinguish between a node that is GRES capable and a node that can act as GRES helper.

the second one is simply an OSPF neighbor of the GRES capable node.

A GRES capable node is a device with hardware based forwarding and two route processors, the event that GRES covers is a switchover of route processors.

Example: GSR or 7600 with two sup720.

GRES capabilities are negotiated between the GRES capable node and its neighbors using protocol specifics TLV.

For OSPF this a special LSA that cannot be passed to other devices : link local scope

if the two devices on the link agree on GRES capabilities this LSA is added to hello messages.

you can see this in show ip ospf neighbors detail.

When the GRES device starts a route processor switchover the new route processor advices its neighbors.

Neighbors hide the fact the device control plane is not operational, and start a timer equal to the one advertised before during negotiation of capabilities by the now restarting node.

This time can be like 120 seconds or more.

During this time they still declare up OSPF adjacencies with restarting node and forward traffic to it according to routing table.

neighbors do not honor the GRES timeout only if topology changes during the time the node is restarting.

Restarting node has to start speak OSPF within the timeout or neighbors will declare it down.

Hope to help

Giuseppe

[vincehgov]

Yes it helped alot.  I also noticed that my 2851 routers do not have the NSF commands available while my 4510R switch does.  The 4510R has dual supervisors so its more clear to me.

Basically, this will only work with routers/switches that have redundant supers..