VMware Cloud Community
WillGreen
Contributor
Contributor
Jump to solution

NIC Teaming and Load Balancing on ESXi 3.5

I want to get ESXi to load balance traffic to my NFS storage, but haven't had any success. My test set up is as follows:

  • Two blades each with two physical NICs dedicated to ESXi kernel ports for NFS

  • Two switches

  • One NIC on each blade is connected to each switch - there isn't any Etherchannel or trunking configured

  • Two NFS appliances on different IP

  • Eight VMs on each blade, four connecting to a data store on each NFS appliance

  • ESXi NIC teaming across the two NICs with the option "Route based on the originating virtual switch port ID"

My understanding was that each VM should use one of the NICs, but that roughly half should use one physical NIC and half the other. Instead, all traffic from all eight VMs goes out of one physical NIC. If I disable the physical NIC it moves over to the second NIC without issue. So they're both working, but it never uses both at once. This behavior is consistent across both blades.

This is limiting throughbut to ~110MB per blade and the storage system can easily exceed this.

I have attached a screenshot of a virtual switch set up.

TIA,

Will

Reply
0 Kudos
1 Solution

Accepted Solutions
weinstein5
Immortal
Immortal
Jump to solution

Are the VMs accessing the NFS datastores directly? If so then yes it should work that way - however if you are using the NFS datastore to house the VM files including virtual disks then the traffic will go out the vmkernel port and if have single vmkernel port then what you are experiencing is expected - the vmkernel selects the outbound NIC based on the virtual port id so if you have a single vmkernel port it will just have a single port id so only a single NIC will be used -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

View solution in original post

Reply
0 Kudos
3 Replies
Amnexi
Enthusiast
Enthusiast
Jump to solution

If you haven't configured any team/bounded group, etherchannel or simmilar, yes, it should work because you have more vNICs than pNICs... (check http://www.vmware.com/files/pdf/virtual_networking_concepts.pdf)

You could check the MAC hash policy to see if the problem is the same for both failover techniques; at least, you will isolate or "extend" the issue Smiley Happy)

Reply
0 Kudos
weinstein5
Immortal
Immortal
Jump to solution

Are the VMs accessing the NFS datastores directly? If so then yes it should work that way - however if you are using the NFS datastore to house the VM files including virtual disks then the traffic will go out the vmkernel port and if have single vmkernel port then what you are experiencing is expected - the vmkernel selects the outbound NIC based on the virtual port id so if you have a single vmkernel port it will just have a single port id so only a single NIC will be used -

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful
Reply
0 Kudos
WillGreen
Contributor
Contributor
Jump to solution

That makes complete sense. ESXi is making all the connections via its one kernel port, so without trunking this will use one interface or the other, but not both. I have confirmed that multiple VMs with NFS connections from the guest OS make use of both links.

Thanks for your help,

Will

Reply
0 Kudos