If no one replies back with ICMP Unreach type 3 code 4 with "Next-hop MTU" then yes http://www.faqs.org/rfcs/rfc1191.html Cheers Alex ----- Original Message ----- From: Keegan Holley To: Alex Cc: Ido Szargel ; juniper-nsp Sent: Wednesday, June 22, 2011 3:27 PM Subject: Re: [j-nsp] BGP MTU Mismatch
I don't think this is correct. PMTU discovery uses probes with the DF bit set, the largest one to make it end to end is assumed to be the MTU. http://en.wikipedia.org/wiki/Path_MTU_Discovery 2011/6/22 Alex <[email protected]> PMTU relies on someone in the path telling you what the MTU is. And there is no way to check if this someone is telling the truth :-( I saw this in complex environments such as BGP over GRE over IPSec tunnel. In this case, the sp-* interface carrying GRE tunnel has to have mtu-discovery itself enabled. Otherwise it will assume 9192 bytes. Cheers Alex ----- Original Message ----- From: "Ido Szargel" <[email protected]> To: "Keegan Holley" <[email protected]>; "juniper-nsp" <[email protected]> Sent: Wednesday, June 22, 2011 9:50 AM Subject: Re: [j-nsp] BGP MTU Mismatch Hi Keegan, What is most likely to happen is that the session established since those are small packets, then the routers try to send the route messages which are large in case you are sharing full routing table, those messages do not arrive to the other side because of the MTU mismatch, thus causing the keepalive messages to be delayed as well, then when the dead interval kicks in the session flaps. PMTU under the BGP group/neighbor should normally be enough to solve it. Regards, Ido. -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Keegan Holley Sent: Wednesday, June 22, 2011 11:38 AM To: juniper-nsp Subject: [j-nsp] BGP MTU Mismatch Does anyone know why a BGP session would constantly flap because of an MTU mismatch. I'm sure it's MTU since that is what fixed the problem. The peering is between a cisco and a juniper and both support PMTU discovery. I would assume any mismatches would be settled by the TCP MSS negotiation or fragmentation (admittedly bad). The peering flapped almost every three minutes on the mark so it never made it past the first dead timer interval. Just curious if someone out there had ever gotten to the bottom of this problem. _______________________________________________ juniper-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/juniper-nsp _______________________________________________ juniper-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/juniper-nsp _______________________________________________ juniper-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/juniper-nsp _______________________________________________ juniper-nsp mailing list [email protected] https://puck.nether.net/mailman/listinfo/juniper-nsp
