[c-nsp] Cisco IGP / BFD Dampening (Suppressing Unstable WAN Links)

Troy Boutso sensible115 at gmail.com
Thu Apr 27 00:52:37 EDT 2017


Hey all

I have a fairly easy one for ya'll (I hope).
I have a network consisting of multiple WAN Links which span all over the
place. By WAN Links I am referring to private leased Layer 2 Circuits with
Ethernet Hand offs.

Each of our client sites has two of these types of links via alternate
providers back to our 2 main "hub" sites.

I am currently running. IS-IS (Level 2) with MPLS LDP and BFD for failure
detections on every single link.
iBGP (MP-BGP) is how I exchange all NLRI between sites... that all works
perfect.


I'll start off by saying 90% of the failures on my network involve one of
these links going down due to some fibre cut or other carrier transmission
issue. This is usually no big deal. As the BFD will detect the fault within
about 2 seconds and traffic re-routes instantly. I usually see a few
packets drop during this time. For what I do. This is completely acceptable.

However yesterday I was thrown a curveball. Rather than a particular link
simply going down. I saw it flap constantly for 2-5 minutes. Which is far
worse than I've experienced in the past. In the past this has happened, but
wasn't as bad (as in wasn't noticable for my customer). My model has stood
up ... until now. I understand what happened. I just don't know the best
way to address the issue and prevent it from being customer impacting.


The flaps i am referring to are NOT link state events. Meaning the
interfaces don't actually change state (the issue is somewhere in my
provider's end). Therefore I don't think "ip dampening" would help. What I
want is something at the BFD, ISIS level. To essentially poison/penalize or
even disable forwarding over link in question for a certain amount of time.
Until stability is maintained. And if the flapping persists, to continue
applying the penalty... (hope you all get what I am saying here)

I looked into "BFD Dampening" which is configured under a template. But is
this really going to give me the desired result? Is there something else
out there that I've overlooked?

If BFD damping is correct. I've got questions about the values :)


Kind Regards
Troy


More information about the cisco-nsp mailing list