[c-nsp] Routing around errors

[Pshem Kowalczyk]

Hi,


On 10 December 2011 09:16, Geoffrey Pendery <geoff at pendery.net> wrote:
> We've seen an old nuisance issue becoming more prevalent lately - at
> dual-homed sites one of the circuits will run errors, but not bad
> enough to drop the routing protocol neighbor (OSPF in our case).
>
> So a site with one good circuit and one error-heavy circuit will
> continue to route traffic over the circuit with errors.  I'm curious
> what solutions are prevalent out there, anybody want to weigh in?
>
> Tune the hellos tight enough in the hopes of dropping the neighbors?
> PfR/OER?
> Alerts from monitoring system followed by manual intervention?
> EEM scripts checking the counters?

Depending on the hardware (and required convergence time) we use the
following methods:

1. G.703 detection with thresholds to monitor number of errors per
second (that's mainly on core links)
2. Ethernet OAM (mainly on PE-CE links)
3. BFD (usually where we can't use the previous two methods)
4. Routing protocols timers/hellos - that's last resort, as the
convergence time is usually pretty bad here.

kind regards
Pshem