Switching

I have EX4200s and EX4500s in a VC configuration, currently running JUNOS 12.3R8.7

Question 1:

Hypothetically, Imagine an EX switch has a port "xe-0/0/1" set to use Jumbo Frames, and a Device also configured to use Jumbo Frames connected to that port.  That device sends a frame that is destined for a second device that does not currently support jumbo frames, and the second device is plugged into another port "xe-0/0/2" on the EX switch which is also set to a regular MTU size of 1500.  

How does the Juniper handle this?  Does it do anything fancy and break up the frames into smaller frames since the jumbo frame is headed to a port that doesn't have jumbo MTU size set?  Or does it simply copy the frame leading to fragmentation?

Question 2 (not likely I'd want to do this, but I'm curious what would happen):

What would happen if I set an MTU to the jumbo size on a port with a device that does not support jumbo frames?  Will the switch learn that the device can only support 1500MTUs, and only send such frames accordingly?

------------

I'm trying to plan a switch to jumbo frame support for our storage pathway for VM data and other scenarios as well, but I'm trying to understand how the mixed environment copes with this.

Thanks a ton in advance!!!

I am not 100% sure of the answer but logic tells me:

Case #1 - the packet is sent as is to the receiver.  Reason is that MTU size is only checked on Ingress, not Egress.  The Receiver can not accept such a packet (per your example) so hopefully it utilizes something like RFC 1191 MTU Path Discovery to tell the sender, try a smaller size, until the sender sends a packet of the size the receiver can handle. Once the sender discovers the receiver can only receive smaller MTU sized packets, it then will fragment the packet into whatever number of proper size MTU packets that the receiver can handle. Fragmentation is NOT done generally performed by routers/switches, but instead is an end device function. 

Case #2 - Since the sender can never send packets greater than 1500, which is less than the MTU size of the interface, the interface does nothing special besides pass the packets, and IF the sender sends bigger, it will also pass.  Interface does no MTU "learning".  Now if ingress packet is greater than the configured MTU for the interface, then the packet is dropped.

I believe this is the way this all should work.  Never tested it though -;)

Question 1: Assuming L2 connectivity between all devices (eg: a single VLAN) the jumbo frame will simply be dropped by the device that is configured with the 1500-byte MTU and the "Oversized Frame" counter on the interface will be incremented.

Question 2: The same as above, except the device will drop the frame instead of the switch.

The thing to remember with Jumbo Frames is that they are only used when upper layer payloads request them, so, for an overly simplified example, take all the switches out of the equation and imagine we have a storage array with jumbo frames turned on, and a server PC with them turned off connected back-to-back.

If a TCP session (say iSCSI) is started up, then the PC (knowing its MTU is only 1500 bytes) will set Maxiumum Segment Size (MSS) to 1460 bytes (to allow for IP/TCP headers etc) and the storage array (even with Jumbo Frames enabled) will then never send a TCP payload larger than 1460 back to the client, which means everything will work nicely.

Network infrastructure in general doesn't initiate connections, so increasing the MTU will not affect transit traffic.

However for some network protocols (BGP and OSPF come to mind) matching MTUs for adjacent peers IS a critical part of operation, so make sure you consider any links that run these protocols (and backup links that may carry these protocols) when making changes.

My recommendation would be:

1. Audit all your network devices to make sure that they support Jumbos first (and what their maximum is)

2. Confirm that if you have mixed vendor environment that the MTU means the same thing on all boxes (on Juniper MTU on an interface includes the L2 header, whereas on Cisco gear it does not)

3. A lot of older equipment may require a reboot in order to change MTUs (Junos does not)

4. Watch out for things like L2 MTU and L3 MTU - in Junos, they are both set independently of each other - for example, a physical port can have an MTU of 9000 bytes, but if it is a member of a VLAN, the l3-interface will have a default of 1500 bytes, which means routed traffic will be forced to this lower value

Thanks for the help guys!