How to use vmnic0 and vmnic1 properly for performance?

[Dave_the_Wave]

Months ago, when I set up a lab using a handful of old Hpe ProLiant G6's (which run ESXi 6.0U3 Hpe version flawlessly), I had only plugged in one patch cable to the two onboard nics. ESXi installs with only a vmnic0. I've even played with vMotion on that setup.

I decided to patch cable the second nic to the switch as well, but wasn't sure what I should do with the vmnic1 that appeared.

The simple thing I did was just add vmnic1 to the vSwitch0, but I have no idea about the "Teaming and failover", I left that default.

I noticed when I click "vmnic0 1000 Full", it would orange highlight the vmk0 and the vm's.

When I click "vmnic1 1000 Full", it would orange only the vm's.

Since vm's are orange on both nics, this means when a client does … rdp/ica into a vm, the host will decide which nic the client goes through?

I'd like a standard set up where host1 will run vm's off the datastore of host2. Because running vm's off their own host's datastore is a no no. Want to keep it simple with all hosts being an ESXi box before I dive into dedicated iSCSI server or vSAN.

I'm mostly interested in performance, and less interested in redundancy, I don't want to do anything production fancy like getting a whole second switch for failovering. If a nic is ever going to die on these 10 year old boxes, they're getting tossed out.

[Rahul Parmar]

Currently, you have configured both the nics in "Route based on originating Virtual port" Load balancing method which is perfect for performance and load balancing.

[Dave_the_Wave]

I understand the VM Network is to be given to as many Physical Adapters as possible, but what about the vmk0 on the Management Network? Should that be on one Physical Adapter as previously illustrated above, or also given to both Physical Adapters as shown below? The following is what happens by default when ESXi is installed with both Physical Adapters patched with ethernet cables.

What would be best practice for more Physical Adapters, like 3 or 4 of them? Is it really necessary to dedicate any for vMotion, when that sort of work is done during work afterhours?

[daphnissov]

It's always a best practice to have redundancy for any and all kernel services (ex., management and vMotion) as well as virtual machine traffic.

Is it really necessary to dedicate any for vMotion, when that sort of work is done during work afterhours?

Your assumption that vMotion is only invoked after hours is generally not the case (or rarely), because DRS uses vMotion as its mechanism and that can occur at any time during the day. For host maintenance and evacuations, however, this might be the case. Even still, it's best to have at least one vmnic dedicated to vMotion if not two.

[Dave_the_Wave]

Can I remove the 'Teaming and failover' Overrides from the 'Management Network' and the 'VM Network', if I am fine with the global settings of the vSwitch0 ?

[daphnissov]

Yes, if you're not actually overriding anything it won't have an effect.