[e-nsp] OSPF with IP-MTU=9000?

[Simon Lockhart]

All,

I'm new to Extreme (I run mostly Cisco networks), and have had a bit of baptism
by fire trying to get everything working.

I've added a couple of X670V's, and a bunch of X460's to our network, all 
running 15.3.1.4 v1531b4-patch1-14.

I've configured a 10G uplink between an Extreme X460 and a Cisco 6500.

Relevant config on the Extreme:

--------------------------------------------------
create vlan "vlan10"
configure vlan vlan10 tag 10
configure vlan vlan10 ipaddress xxx.xxx.129.182 255.255.255.252
enable ipforwarding vlan vlan10
configure ip-mtu 9000 vlan vlan10

enable ospf
configure ospf add vlan vlan10 area 0.0.0.0 
configure ospf vlan vlan10 priority 0
--------------------------------------------------

Now, every 100 minutes, OSPF adjacency between the Extreme and the Cisco drops,
and then re-establishes. This co-incides with an OSPF LSA flood.

On the Extreme, I get the log messages:

08/06/2013 15:47:08.39 <Warn:ospf.event.PktLenInv> Packet is too short (ipHdr=20, buf=8192)
08/06/2013 15:49:06.67 <Warn:ospf.event.PktLenInv> Previous message repeated 25 additional times in the last 114 second(s)
08/06/2013 15:49:12.80 <Noti:ospf.neighbor.ChgState> Changing the state of neighbor rtid xxx.xxx.129.24 ipa xxx.xxx.129.181 to state = INIT.

and on the Cisco:

Aug  6 15:49:06.654 BST: %OSPF-5-ADJCHG: Process 1, Nbr xxx.xxx.129.227 on TenGigabitEthernet1/4 from FULL to DOWN, Neighbor Down: Too many retransmissions


If I change the IP MTU back to 1500 on each end, the problem goes away.

I've got an SR open with Extreme TAC, but don't seem to be getting anywhere
fast with them - they're convinced there's an MTU mismatch or something.

Has anyone else seen this before? Does the "buf=8192" in the Extreme error 
message alude to the fact that the Extreme won't work with a packet bigger than
8192 bytes? I'm currently testing with an IP MTU of 8000 to see if it makes
a difference, but have to wait 1hr40 to see if it does :(

TIA,

Simon