OSPF recalc

Take a look at

It will likely answer your questions and the SPF throttling could help mitigate issues with SPF recalculation happening too often.

Cisco da Gama

Thanks for the feedback. The link is interesting and may help.

I would still like to understand when a OSPF router stops forwarding traffic due to receiving LSAs. I'm looking for a theoretical answer that can be used as a base line.

an article that explains how the recalc process works in various setups.

this article explains the thinking behind the timers in IOS that limit when the SPF calculation can happen - this is there to limit CPU use and avoid starving the other router processes of processor time, but the side effect is that the router doesnt immediately react to some events.

eg. you have an Ethernet link between 2 routers, with a full adjacency. 1 port goes down. There is an interface timer which doesnt report the event to OSPF to force an event and recalc until some time elapses - so the router sends packets to the down route in the meantime.

once that timer expires the router can recalc routes fairly quickly - but it depends on no. of routes, neighbours, area size and router performance, and this is square law scaling issue. this is where the answer is going to be "test it and see what happens"

So the answer to your Q about what does the router do to routes during the Dykstra - i dont think the standard defines this (or different routers work differently and still meet the standard).

I think the default on the Cat 6 in the lab at work running 12.2.18? is that it will invalidate routes pointing out the affected interface, but others stay up during the recalc (since it only takes a mSec or so, without some serious test gear it is difficult to tell). This may or may not generalise to all IOS flavours and hardware platforms - i suspect this has been tuned in recent IOS releases to minimise the effect of rerouting on time sensitive protocols, like VoIP.

Certainly a lot of Cisco recommendations about high availability stress using equal cost multipath to minimise the forwarding "hit" when rerouting happens - and that wouldnt make sense with OSPF unless routes out of other interfaces stay up during a partial recalc. There is also a recent feature (in 12.4?) called incremental SPF designed to limit the scope of a Dykstra so what recalc there is happens faster.

However - in a lab setup the delay before OSPF triggers the recalc dominates traffic loss - so the other answer is in my case it doesnt matter in practice.

>

simple answer is: if the router is constantly receiving changes to the network topology it cannot come up with a stable picture of the topology from which to create the routing table. the practical limit would be roughly proportional to cpu power.

Thanks for the replys.