OSPF Convergence time

aravindhs · ‎04-13-2007

Hello Experts

I have a basic question about routers convergence times.

In campus networks, We know that it is better to go for a layer3 backbone instead of a layer2 backbone for one of the major reasons of delay in STP stabilization times which the routing protocols' convergence times can achieve.

Now to my basic question ..

Time for a normal 802.1D STP to converge is, lets say 45 to 50 seconds. (20 maxage + 30 Fwd_Delay)

Lets consider OSPF with a hello time of 10 s and deal interval of 40 s.

This means that a router B (connected to Router A) will only bounce the neighbourship with A and purge the routes learnt from A only after 40 secs of not hearing any hellos from A and run SPF again.. So it can't be that way, can it ? What basic idea have I not understood here ?

Thank you !

cheers

arav

chrihussey · ‎04-13-2007

Your assesment of the convergence times is correct to a degree. But consider the following:

If you had two routers/switches connected via redundant L2 links. One link would forward the traffic the other would be in a blocking state to prevent an SPT loop. If the active one failed, then the 45 second calculation time would be required to start forwarding on the other link.

Now if you had two router/switches connected via separate L3 links with OSPF routing enabled across them. Both links would be active and load share. If a link failed, the path is immediately purged from the link state database and all traffic reroutes to the other path in seconds. There is no time spent on hellos or dead intervals. The same would apply on a larger scale if implemented correctly.

Also, one of the biggest advantages to a L3 backbone as opposed to L2 is pure stability in the sense that broadcast traffic is better contained and the chances of a spanning tree loop occuring are greatly dimished as well as if one did occur it would not affect the entire network.

aravindhs · ‎04-16-2007

Thanks Chris !

I understand its indeed better to do L3 routing as opposed to L2 core connections.

I was just keen to find out what other ways we could compare the convergence time of OSPF with a standard Dot1D STP convergence times.

cheers

Arav