3850 Global MTU: Increase without rebooting stack?

[2ndcongress]

Is it possible to change (increase) the Global MTU value on a 3850 stack without rebooting the stack?  We're finding mixed information on this subject.

We have a production core switch stack of 3, WS-C3850-24XS switches on version 03.07.04E.  The Global MTU for the stack is currently set to 9000.  We need to trunk an iSCSI VLAN from a 6500 which is already configured with an MTU of 9216.  The iSCSI SAN and VMware hosts (on the 6500) are all configured with an MTU of 9000.  So it would appear that we need to raise the MTU value on our 3850 core to a value higher than 9000 to support iSCSI hosts on the 3850 (same VLAN).  Rebooting the 3850 switch stack is not really an option unfortunately.

Does anyone have experience with changing the MTU on a 3850 running 03.07.04E or newer?

Thanks!

Core3850#show system mtu
Global Ethernet MTU is 9000 bytes.

WS-C3850-24XS

Cisco IOS Software, IOS-XE Software, Catalyst L3 Switch Software                  (CAT3K_CAA-UNIVERSALK9-M), Version 03.07.04E RELEASE SOFTWARE (       fc1)

[Philip D'Ath]

It requires a reboot alas.

[akemmy]

Hello,

My lab 3850 is running 

Switch Ports Model SW Version SW Image Mode
------ ----- ----- ---------- ---------- ----
* 1 56 WS-C3850-48U 03.06.06E cat3k_caa-universalk9 INSTALL

I boot the switch with no "system mtu" in the config, so it is running default 1500

I can change the MTU via command "system mtu 1650"

I have 2 laptops plugged into the switch on the same VLAN, with jumbo frames enabled on their ethernet interfaces, running "ping -t -f -l 1550" - 1550 size frames with the "do not fragment" bit set.

Before I change the switch MTU, the pings from one laptop to another fail.

As soon as I change the switch MTU with "system mtu 1650" - without rebooting - the pings succeed.

It seems the MTU settings take effect immediately, with no reboot required. 

If this MTU setting is dynamic in practice, why the recommendation to reboot?

       Stability? Performance?

       It's just the recommended procedure, could cause support issues if you don't?

The nearest I have found online is that "the ASIC gets reprogrammed on reboot to support the new MTU size". Okay, but if the feature works without a reboot, exactly what is being reprogrammed, and what is the implication of not doing it? Is there a risk of running out of buffer space, memory leaks etc?

Asking as the production environment (3850 stack) we are looking at is both intolerant of instability and intolerant of outages to reboot the switch stack.

Would be happy with a reply such as "even if the command takes effect dynamically, the recommendation is to reboot to prevent the possibility of future network instability".

Thanks,

Andrew