Hey all, have a couple of design questions. I have one group on campus asking for "jumbo frame support" for their servers. In our campus network, each building has a router and multiple switches. That building router uplinks to distribution.
I asked the group to clarify if they want server to server same vlan jumbo, or servers in different vlans (on the same building router) to exchange jumbo packets. I don't have that answer yet, but I would like to clarify my limited understanding of mtu.
In this case we are talking about a Cisco 3850 L2 stack that connects to a Cisco 3850 L3 for building routing. If the design goal is layer 2 same vlan only jumbo, then I would go to each interface and issue an MTU XXX command, yes? But the only command I can find on the 3850 is system mtu xxxx. This changes all L2 and L3 mtu. Does this include the ip mtu? Could this not break OSPF, etc if done on a router?
The 3850 L2 stack is pre-production, so I applied the system mtu 9000 command to see what would happen. Right after I did a sh ip int | i MTU and a sh int | i MTU and in both cases the MTU had changed to 9000. I notice there is an ip mtu xxxx command under the SVI, but hasnt that already been changed with the system mtu command?
This should help illustrate:
slc-sw-1111-b(config)#system mtu 9000
Global Ethernet MTU is set to 9000 bytes.
Note: this is the Ethernet payload size, not the total
Ethernet frame size, which includes the Ethernet
header/trailer and possibly other tags, such as ISL or
802.1q tags.
slc-sw-1111-b(config)#do sh ip int | i MTU
MTU is 9000 bytes
slc-sw-1111-b(config)#do sh ip int brief | e unass
Interface IP-Address OK? Method Status Protocol
Vlan480 172.16.9.6 YES NVRAM up up
slc-sw-1111-b(config)#int vlan480
slc-sw-1111-b(config-if)#ip mtu ?
<832-9000> MTU (bytes)
slc-sw-1111-b(config-if)#
As you can see there is only one ip interface on this L2 switch. If I look at the ip mtu command under that interface after the change, it says I can go up to 9000. OK, that makes sense, but it seems that the ip mtu was already changed with the system mtu command. Would this not break any OSPF on the spot? If I have to change all my ip mtu to 9000 just so that I can then selectively have some running at 9000, isn't that problematic? Maybe I am missing something.
Also, on a related side note:
I am wondering if the following situation would create ip framgentation. It was suggested in a meeting recently that we should enable jumbo frames on our building to distribution uplinks, but not on building vlans unless specifically requested. Would this not cause fragmentation? We would have a point to point with 9000 mtu on each side, but then when traffic reaches a router (L3 3850) bound for say a host in vlan 420 in that building, would the 3850 not have to then fragment since the destination vlan is not configured for the larger ip mtu?
Sorry for such a long post. I appreciate your taking the time to think about this and provide any responses.
