Change iSCSI to Jumbo Frames

[alainrussell]

Hi, I've been reading about jumbo frames and am seeing some odd performance under load on our iSCSI SAN that leads me to think this will help us. When recreating the vmkernels and switches, will ESXi reconnect to the datastores for the current VM's we have ok, are there any gotchas with this that I need to consider?

Thanks

Alain

[golddiggie]

You'll need to enable jumbo frames on every piece of hardware from the host to the SAN. That includes any switches, or at least the ports used on the switches, that are configured for the iSCSI traffic. I would advise doing this during an off hours/maintenance window since you might need to reboot the SAN in order to get jumbo frames fully enabled. If the SAN won't require a reboot, then just make sure you have all your processes lined up.

It also depends on which version of ESXi 4 you're running. If on 4.1 you might be able to just enable jumbo frames on the vSwitch configuration on the host(s) side. If 4.0, you'll want to recreate the vSwitch via CLI so that you can enable jumbo frames on it.

If it was me doing this, I would make sure I had the steps down for each piece of hardware in the chain, before moving forward.

Out of curiosity, what is your SAN and what do you have for switches in the environment?

[royberk]

When changing iSCSI to jumbo frames you also need to change the config on the ESX host in the command line to enable on the vSwitch:

esxcfg-vswitch -m 9000 "vSwitch name"

I would do this first before enabling on the SAN side, and make sure your switching has jumbo frames enabled as well.

Enabling it on the vSwitch may disconnect the VMs connected, but I dont believe it will have an affect on the datastores connection.

Lets see what others may have to say

[alainrussell]

Thanks, the SAN is an MD3000i. We're running 4.1 so can I do the changes without using the command line? I haven't seen that referenced anywhere else - most of the info I've read online states its a delete/recreate on the vmkernel & switch.