I second paul, What is the use of the routes learnt when we know forwarding is going to fail, And secondly the forwarding will fail selectively only for the large data-packets that makes the troubleshooting more difficult
Regards Prasanna -----Original Message----- From: Paul Wells [mailto:[EMAIL PROTECTED] Sent: Thursday, May 29, 2008 3:01 AM To: Acee Lindem Cc: Prasanna Kumar A.S; OSPF List Subject: Re: [OSPF] What is the use of MTU field in DD packet Hi Acee, That was before my time, so I'll defer to your recollection about how OSPF MTU checking came to be. The section of 2178 below does seem to give equal weight to both the control and data plane benefits of MTU verification however. This still leaves the question of what to do for af-alt when routing address families other than IPv6. It seems that there are two cases of interest in deciding which MTU to advertise in the DBD packets: 1. IPv6 MTUs match, but IPv4 MTUs differ. 2. IPv6 MTUs differ, but IPv4 MTUs match. In the first case I don't think we're doing anyone a favor by installing routes in the IPv4 RIB that will be unreliable due to a MTU mismatch. In the second case OSPFv3 flooding and synchronization may be compromised. A side effect of this may be that the adjacency never forms, or having formed may later fail. Short of resorting to LLS or some other way of communicating both MTUs it seems we have to pick one or the other. I'd like to propose that we use the DBD packet to communicate the IPv4 MTU when routing that address family, and use the IPv6 minimum MTU of 1280 bytes for OSPFv3 protocol packets. Thanks, Paul Acee Lindem wrote: > Hi Paul, > > On May 28, 2008, at 2:05 PM, Paul Wells wrote: > >> Hi Acee, >> >> I disagree about the "original intent" of the MTU field. As I see it, >> it's function is to prevent an OSPF adjacency from forming over a link >> where the endpoints disagree about the link MTU. We do this primarily >> to prevent the data plane from using a link that will drop packets >> sent to a system with an MTU smaller than ours. > > I happen to remember the discussion of this problem on the OSPF list and > this was not the primary motivation. There were lots of problems with > bridged heterogeneous LANs with mismatched MTUs (ethernet, FDDI, token > ring, and the worst of all technologies - ATM emulated LANs :^). > Adjacencies would come up fine initially but the exchange process would > hang indefinitely when they were restarted due to the router with the > larger MTU having a larger database and trying to use full DD packets. > Unfortunately, the OSPF list was hosted on a server at Microsoft > Corporation in those days and I don't have access to archives. Here is > some text from RFC 2178, appendix G: > > G.9 Detecting interface MTU mismatches > > When two neighboring routers have a different interface MTU for their > common network segment, serious problems can ensue: large packets are > prevented from being successfully transferred from one router to the > other, impairing OSPF's flooding algorithm and possibly creating > "black holes" for user data traffic. > > This memo provides a fix for the interface MTU mismatch problem by > advertising the interface MTU in Database Description packets. When a > router receives a Database description packet advertising an MTU > larger than the router can receive, the router drops the Database > Description packet. This prevents an adjacency from forming, telling > OSPF flooding and user data traffic to avoid the connection between > the two routers. For more information, see Sections 10.6, 10.8, and > A.3.3. > > On the other hand, once the MTU checking was implemented, I believe data > plane MTU consistency has been purported as a benefit. If we used the > IPv4 MTU in the IPv4 address database exchanges, we could still have an > IPv6 MTU mismatch. One could depend on the unicast IPv6 address family > for this checking but, heretofore, we've kept the instances independent. > > Thanks, > Acee > > >> >> While OSPFv3 certainly needs to know the IPv6 link MTU when building >> it's packets, this information should be available locally without >> reference to the MTU field in the DBD packet. >> >> So, I would argue that in af-alt the MTU in the DBD packet should be >> for the address family we are routing, not IPv6 in all cases. >> >> Regards, >> Paul >> >> Acee Lindem wrote: >>> Hi Prasanna, >>> On May 28, 2008, at 8:18 AM, Prasanna Kumar A.S wrote: >>>> Hi >>>> I just wanted to understand what the primary use of exchanging MTU in >>>> DD packets and doing MTU-check is? Is it only for the control plane or >>>> is it for the DATA-plane? >>> Control-plane - when sending DD, LSR, and LSU packets, OSPF will >>> attempt to send as many LSA headers or complete LSAs as will fit in >>> a maximum sized packet. >>>> Why I am getting this doubt is, in draft-ietf-ospf-af-alt-06.txt >>>> doesn't >>>> specify which MTU we should use while exchanging the DD packet for the >>>> ipv4-unicast or ipv4-mutlticast Address-family, is it ipv6-mtu or >>>> ipv4-mtu? >>> We have this clarified in the an update which we post soon. Since >>> this is OSPFv3 which using IPv6 for transport, you always use the >>> IPv6 MTU. >>> Thanks, >>> Acee >>>> >>>> Regards >>>> Prasanna >>>> _______________________________________________ >>>> OSPF mailing list >>>> [email protected] <mailto:[email protected]> >>>> https://www.ietf.org/mailman/listinfo/ospf >>> _______________________________________________ >>> OSPF mailing list >>> [email protected] <mailto:[email protected]> >>> https://www.ietf.org/mailman/listinfo/ospf > _______________________________________________ OSPF mailing list [email protected] https://www.ietf.org/mailman/listinfo/ospf
