[c-nsp] OSPF per-prefix LFA

[Dan Peachey]

On 28 May 2015 at 14:43, Adam Vitkovsky <Adam.Vitkovsky at gamma.co.uk> wrote:

> Hi Mohammad,
>
> I have never had to tune IGP to get sub 50ms LFA failover times.
> The failover times have nothing to do with IGP.
> Actually the switchover itself (from primary to backup path) is done in
> couple of usec (microseconds).
> So what you need to fight against is actually the time it takes the HW to
> realize there's no light on the link.
> As it was suggested already tuning the "carrier-delay down" to 0 is an
> absolute must.
>
> Yes tuning:
> timers throttle lsa,
> timers lsa arrival,
> timers pacing flood
> - is vital to propagate the information across the backbone as fast as
> possible so BGP can switch to backup NH ASAP if primary NH is unreachable.
> But that's a different story altogether.
>
> I think it's really hard to tell what's actually going on in virtual
> environments.
> Looks like in your case LFA is really slow so SFP kicks in sooner than LFA
> that's why tuning SPF got you better results.
> You might want to check how your virtual environment reports link failures
> (at what time intervals).
> Is LFA actually supported on CSR?
>
> adam
>

I suspect it's just the RIB route that is being determined by the IGP
re-calculation (and therefore SPF timers) and actually the FIB route is
changing instantly as expected.

I'd assumed that the RIB route would change along with the FIB route but it
sounds like stale route is kept until SPF calculation is run and new best
path is put in RIB (which is already in FIB anyway). Also I guess you would
want to calculate another LFA path too (if available). It makes sense now I
think about it.

To be honest I always tested this with tuned SPF timers so never noticed
that the RIB didn't update instantly after link down.

Cheers,

Dan