Aggregating links from Virtual Machine to Distributed Switch

[dockeradz]

I'd like to create a LAG using LACP from a VM to a distributed switch.  When I set this up it works fine to a standard switch but does not work to a distributed switch.  Is there a way to enable LACP on a distributed switch or is that not possible?

[virtualg_uk]

I thought LACP was only possible with a vDS.

Regardless, you'll need to configure via the Web Client. Please see https://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=20342...

[dockeradz]

This sounds like connecting a host to a physical switch.  I am trying to connect a VM to the vDS using LACP.

[MKguy]

A LAG/LACP between VMs and (d)vSwitches is neither supported, nor is it needed, nor makes it sense. This is just asking for blackhole forwarding and physical switch MAC flapping.

LAG/LACP is only supported between physical switches and vSwitches.

If your argument is that you want more bandwidth than a single vNIC can deliver, then this is already a moot point. You already can achieve higher bandwidth on a single vNIC than the link speed it emulates, because physical signaling limitations do not apply to a virtual NIC.

Put 2 Linux VMs with a single vmxnet3 vNIC on the same host and same port group and run iperf or some similar benchmark between them. Depending on your CPU performance you will get 20-30gbit/s over the emulated 10gbit/s link.

Likewise, if you have a LAG of multiple 10gbit/s physical uplinks on a vSwitch then an attached VM can exceed it's single 10gbit/s virtual link bandwidth over the physical network as well.

Read more about this here:

VM Guest Bonding

Maximum traffic for a vmxnet3

Guest-to-guest communication on the same ESXi host

[dockeradz]

My guest OS (OSX) does not have a VMXNET3 driver.  This is the reason I'm trying to bond multiple nics.  My confusion is that it works on a standard vswitch but not on a distributed switch.  I thought maybe I missed something.

[MKguy]

The same principles also apply to other adapters like the e1000, which only presents a virtual 1gbit/s links but it can still exceed that as well if you have 10G or multiple 1G links in a LAG on the physical switch side.

I suppose things working on one side is just a coincidence and has to do with physical configuration and/or the dynamic vNIC/physical uplink mapping.