I have two vSphere hosts (e.g., H0 and H1) with multi-port NICs connected to the same managed switch. I want the VMs on both hosts to have the highest network throughput.
1. On the switch, should I create two LAGs (e.g., LAG0 and LAG1); where LAG0's port members are connected to H0 and LAG1's port members are connected to H1?
2. Is it necessary to put LAG0 and LAG1 on the same VLAN for performance?
3. Does vSphere support LACP?
Hello,
You can achieve great performance by having your traffic split between more NICs - VM traffic, Storage and vMotion comes to mind but the gap slowly closes with the advent of converged network adapters.
Personally, I do not see the advantage of ESXi hosts with LAGs outweighing the increased complexity they might introduce in most practical cases. Take a look at these articles explaining some pros and cons on this debate:
http://wahlnetwork.com/2014/01/13/vsphere-need-lag-bandaids/
http://wahlnetwork.com/2014/02/05/revenge-lag-networks-paradise/
Adding to the previous message, LACP is only supported on distributed vSwitches. The standard vSwitch only supports static etherchannel LAG with source/dest IP hashing.
The dvSwitch LACP implementation in ESXi 5.5 supports a large number of different load balancing modes, see these articles for details:
http://wahlnetwork.com/2014/02/19/exploring-enhanced-lacp-support-vsphere-5-5/
http://wahlnetwork.com/2014/02/19/exploring-enhanced-lacp-support-vsphere-5-5/2/