OSPF hello timer

Are all of the OSPF speakers Cisco ?

What version(s) of IOS is running on the 6500 ?

Have you looked at using BFD ?

"... and Sun load balancing switches..." I think it's the N2000 series.

Nope. I don't think the Suns support it and given how new it seems to be I'd be more than a little chary about it anyway. I'm clearly getting conservative in my old age... :-)

Sam

given OSPF speakers from different vendors then OSPF hello/dead timers are the knobs to use.

are the GE links point-to-point or GE broadcast (ie. more than two OSPF speakers) ?

That's what I thought. Have you (or anyone else who's listening in) any experience of sub-10-second hello timers? Any gotchas with using 1 second? 2 seconds?

Broadcast. It's a VLAN with redundant links spread across two sites. At present there'll be two Ciscos and two Suns involved.

Sam

I have used 1 second hellos on GE interfaces but point-to-point so usually detect physical link failure.

You may also want/need to look at OSP exponential SPF backoff feature

router ospf 1 timers throttle spf 1 5000 10000 end

sh ip ospf Routing Process "ospf 1" with ID 3.3.3.3 Supports only single TOS(TOS0) routes Supports opaque LSA Supports Link-local Signaling (LLS) Supports area transit capability Initial SPF schedule delay 1 msecs Minimum hold time between two consecutive SPFs 5000 msecs Maximum wait time between two consecutive SPFs 10000 msecs Incremental-SPF disabled Minimum LSA interval 5 secs Minimum LSA arrival 1000 msecs LSA group pacing timer 240 secs Interface flood pacing timer 33 msecs Retransmission pacing timer 66 msecs Number of external LSA 0. Checksum Sum 0x0 Number of opaque AS LSA 0. Checksum Sum 0x0 Number of DCbitless external and opaque AS LSA 0 Number of DoNotAge external and opaque AS LSA 0 Number of areas in this router is 0. 0 normal 0 stub 0 nssa Number of areas transit capable is 0 External flood list length 0 IETF NSF helper support enabled Cisco NSF helper support enabled Reference bandwidth unit is 100 mbps

Thanks. I'm trying to work out whether there are any implications for the LSA retransmit-delay - is there any point in dropping that as well?

OK, that seems to have made it into 12.2()SX (it's not in mainline

Sam

this needs you to change the interface default timers - on a Cat6k these give you a 5 sec delay before the switch reacts to a "down" interface.

agreed - 1 mSec is pushing it, and the other timers involved mean 10s of mSec anyway to react.

it depends where you expect "faults" to come from. If the link layer 1 fault detect is going to kick off a Dykstra, then interface timing is the key.

if "dead" timers wont normally detect a problem, then you can relax those and use as the fallback for all the unusual wierd failure modes.

there are a bunch of recommendations about this kind of tuning that may be useful in CCO.

So 100s of ports and 1 sec hello is a recipe for disaster....

you can go to sub second hellos (i think you specify how many per sec, and you get a 1 sec dead timer by default).

finally Cisco support a new point to point "dead detect" protocol, that may mean you dont need to be so aggressive with OSPF timers - BFD or black hole forwarding detection?

SPF exponential setting of 1 ms say that for the first "failure", SPF is to be executed immeditely following the OSPF neighbour adjacency loss. This is a GOOD THING. For the second failure wait longer ( 5 seconds to execute SPF) and for 3rd and supsequent wait even long ( ie. there is a flapping neighbour). This is a very good feature so you may want to make use of it in other oplaces in your network

I would lower the OSPF hello timers first to see if the resulting convergence improvement is satisfactory.

If not then also make use of the SPF exponential backoff feature