iSCSI with JumboFrames

[RLI2013]

Hi all

I do not want to start a discussion about "to JumboFrame or not" but if you go for JumboFrames on an iSCSI storage Network I have my concerns about VMwares best practice.

In all papers I found, the recommendation for configuring JumboFrames was Setting the vSwitch to MTU=9000 and the according VMKernelPort (VMK) as well to MTU=9000. My packet would travel from an ESX Point of view like this:

Esx -> VMK (mtu9000) -> vswitch (mtu9000) -> physical iSCSI Switch (Support for JumboFrames enabled: means max mtu=9216) -> storage NIC (mtu=9000)

So the reason my physical Switch has MTU=9216 is to accommodate Overhead. Why this Setting is not recommended for the vSwitch on the ESXi as well? I would rather configure it like this:

Esx -> VMK (mtu9000) -> vswitch (mtu9216) -> physical iSCSI Switch (mtu=9216) -> storage NIC (mtu=9000)

What are your thoughts? Any Feedback more than welcome.

[RLI2013]

Max. configurable MTU size on a vSwitch is 9000. So even in theorie it could make sense, there is no way to implement that...

I guess on the vSwitch level there is no overhead and that for the configuration with MTU=9000 is fine. But just a guess...

[vasan22in]

Jumbo Frames details,

Enabling Jumbo Frames on virtual distributed switches (1038827) | VMware KB

Jumbo Frames with VMware ESXi – Edge Cloud  ....

[RAJ_RAJ]

Hi ,

If  you have enabled jumbo frames on the  physical switch then on ly you will get the full benefit of increasing MTU on the vSwitch level

[RParker]

If  you have enabled jumbo frames on the  physical switch then on ly you will get the full benefit of increasing MTU on the vSwitch level

Wasn't his question.. did you see what he asked.. is WHY 9216 is the optimum size to allow for extra payload but 9000 is the MAX setting.. that's ALL he wants to know.  In order to get the switch to allow for 9000 packet size payload it's not utilizing 100%..

[RParker]

Esx -> VMK (mtu9000) -> vswitch (mtu9216) -> physical iSCSI Switch (mtu=9216) -> storage NIC (mtu=9000)

In theory it makes sense however I NEVER seen 9216 packet size mentioned before you probably have to allow the switch (network switch not ESX host) to allow for this and since 9000 is the industry standard no one wants to deviate just for .97% change.. how much is that TRULY going to make a difference?

This is why we impose STANDARDS everyone has a tweak here and there but we have to draw the line someplace.

[RAJ_RAJ]

Hi ,

You can enable but if Jumpo frames not enabled on physical what is the use of enabling on vswitch level .

Example - If you are using HP VSA  as storage and as per vensor it is recommenced use 9000 MTU for iscsi but if it is not enabled on physical switch there is no use