cogent update suppression, and routing loops

hi. relatively new cogent customer. is what i've stated in my subject line
kinda standard fare with them?

i've discovered that when i advertise a /24 from inside a larger /22 to XO,
(who peers with cogent), and then pull the /24 some time later, that cogent
holds onto the /24 and then bounces packets around in their network a bunch
of times for upwards of 8-10 minutes until they finally yank it. this
effectively blackholes traffic to my /24 for anyone that is using a path
thru cogent.


it's been a bit of a frustrating experience talking to their noc to
demonstrate it, but i'm able to duplicate it on demand. even pushing routes
using their communities to offload the circuit takes forever to propagate
even on their own looking-glasses.



First time I'm seeing it, and I've been a Cogent client for quite a while.

Have you tried getting in touch with their NOC yet? They're one of the most responsive in the industry.

as stated, yep. i was on the phone with them for over three hours yesterday.

i still advertise the aggregate as a backing route. one reason i might like
advertising a /24 is (usually) it's a nice way to gently attract return
traffic down a certain path so i can do maintenance on the other side.
plenty of other ways to do this, i know (prepending, communities, etc).

Perhaps related, I made an as-path prepend change on a route advertised to Cogent yesterday and noticed it took about 10 minutes for that change to propagate throughout Cogent's network and be visible on route-views. A moment later, I did the same thing with another carrier, and got the expected nearly instant gratification. Perhaps they've got a bunch of routers configured with minimum-advertisement-interval to batch the BGP updates and if you're unlucky with the timing, it can take a while for routes/changes to percolate through their network? Imagine driving down a long road, where every traffic light turns red just before you get to the intersection. Wait a minute here, wait a minute there...

I've had the exact same thing happen.
Recently, I removed a bunch of /24's that were member of a larger /20. We
use to advertise the /24's to certain carriers for traffic engineering. I
saw the exact same issue you're describing.
I chocked it up to Slow BGP in their core. For instance, I removed the
route from my session with them in Orlando. It would get removed from
Orlando, But still exist in Jacksonville. And basically route loop coming
in from the north. It'd land in JAX, And get forwarded to MCO. Which would
then forward it back to JAX. 20-30 seconds later, it would get removed from
JAX, and I'd see the same behavior between JAX and ATL. It took a 4-5
minutes in my experience for the route to finally leave the Cogent
I lessened the blow by making Cogent the first peer I'd remove the /24
from. Leaving the other /24's out there via the other carriers. So only
those that chose cogent as the most direct route felt the unintentional
blackhole. Other carriers removed almost instantly as expected.
Nick Olsen
Network Operations (855) FLSPEED x106

Maybe running into some non-standard BGP Advertisement intervals inside
Cogent's network? Might be running a lot of sub-ASes inside the
confederation, and having to wait multiple advertisement intervals for full

circling back on this, i guess my case with cogent has been escalated to vp
engineering, and i've had a few people reply on and off list citing the
same problems. i encourage you to open up cases to help demonstrate further
examples (ie: it's not just me!)

thx everyone.