On 11/23/2010 5:13 PM, Eric Oosting wrote:
On Tue, Nov 23, 2010 at 3:00 PM, Lobo <[email protected] <mailto:[email protected]>> wrote:On 11/23/2010 6:14 AM, Mark Tinka wrote: Can't think of why this would be an issue. The longest distance we have today between two iBGP neighbors is 160ms (and soon, the farthest we'll have will be about 230ms), and that has no problems at all. I'd suspect MTU issues here. Cheers, Mark. It's looking like it somehow might be related to that. For a test, we set the mtu on the toro-router2's interfaces back to 1500 from 9216. After that was done, the sessions to the vanc routers both came up! Strange because we didn't change any of the mtus on the vanc routers.....they're all still 9216. One thing I didn't mention before was that all of the interfaces on these routers had jumbo frames turned on. The cloud between toro and vanc though is mostly limited to 1546 but this has never posed a problem before with the previous IOS versions.The default MSS in older IOS was something in the 500s, so even if you had a large MTU on the interface and the path would only accommodate 1500ish you'd still be OK. It could be that your newer IOS either went to a larger default MSS that could take advantage of the 9216 MTU and there was no PMTUD, or PMTUD is busted in between so the MTU squeeze between the two routers isn't detected.-e Jose P.S. I've opened up a TAC case as well to see if they can figure something out. _______________________________________________ cisco-nsp mailing list [email protected] <mailto:[email protected]> https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
And we have a winner! I disabled path mtu discovery for both of the remote neighbors and after clearing the sessions they both came up and have been stable since last night.
You're right that it appears that path mtu discovery must be broken or something in this version of IOS because when I look at the MSS for the neighbors they still show as 9176 (9216 - 40) where it should show something closer to the 1506 mark.
Thanks for the tips everyone! I'll let our TAC engineer know about these findings as well.
Jose _______________________________________________ cisco-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
