Hope this isn’t a daft question…..
By default the MTU size in VOSS is set to 1950. In the topology we are using at the moment there is a L3 connection between the new VOSS switches and EOS S8’s. This is configured with a single VLAN configured at both ends as tagged, and an IP address and OSPF enabled between the two.
The interface on the S8 was showing input errors, of which setting the S8 MTU to 1950 to match the VOSS switch stopped the errors.
Not sure what the best thing to do here is i.e. set the MTU on the S8 to 1950 or maybe the VOSS port to 1500?
My thoughts are to match the MTU on the S8 to so as not to force packets to fragment if I did it on the VOSS side.
The question is if I’m introducing MTU sizes of 1950 into the S8 network to general traffic passing between the two am I just pushing the problem elsewhere into the network?
Many thanks in advance?
Best answer by Ludovico Stevens
Switches won’t fragment, only software-based routers can do that.
If a packet is larger than the sys MTU set, a VOSS switch will simply drop the packet.
Normally end-stations use a default IP MTU of 1500 bytes. So the largest Ethernet packet you will get is 1522 (if Q-tagged) and 1518 (if untagged). In the data center for backups you would maybe set an IP MTU of 9000 on the server NICs, but this traffic would not usually spill into the Campus.
Fabric technologies, use an overlay, which uses an ecapsulation. VOSS Fabric Connect uses a MAC-in-MAC encapsulation so you will get oversized (meaning > 1522 bytes) packets on NNI links only, which is why on VOSS the default MTU is 1950 (a safe value higher than 1522). If you set the VSP sys MTU to 1522, Fabric Connect won’t work anymore, so you probably don’t want to do that. But you can set the MTU to 9600 and Fabric Connect will work with jumbo frames as well. VOSS only allows these 3 values: 1522, 1950 & 9600.
I don’t know the S8 well enough. But if the MTU setting is for an Ethernet MTU (as opposed to IP MTU), then 1500 is too small; it needs to be 1522 minimum I’d say.