3

I'm setting up a 2-host VMWare 5.1 Cluster to use a vDS (Virtual Distributed Switch) with LACP enabled, over 4 NICs (2 per host).

In times past, when I've used VMWare Standard Switches with Static LAGs, each host got their own LAG Group on the physical switches. So for the same scenario (but not using LACP or vDS) I would have two LAG groups on the switch side for the 2 hosts. In theory each LAG group would span just 2 switches if I wanted switch redundancy.

All the resources I could find on internet seem to say that vDS in VMWare 5.1 is limited to one LACP group per vDS. Is that saying that on the physical switch side, you define 1 LACP group that handles 4 ports together as one unit? Or do you still have 1 LACP group per host (so 2 LACP Group in my scenario)?

If you have to define 1 LACP group on the switch side, does that mean that as you add each new host to the same vDS, you keep having to grow that one LACP group to add ports potentially from many different switches?

Crispin Bivans
  • 33
  • 3

1 Answers1

2

You create one dvSwitch. You add the hosts to that dvSwitch. You then create an LACP group on the switch for each server connecting to the switch. Assign trunks/vlans appropriately.

Tell us about the switches In use, and I can give more specifics.

ewwhite
  • 197,159
  • 92
  • 443
  • 809
  • From LLDP, I can see it's a Cisco N5K-C5596UP running Nexus OS 5.1. All the existing ports claim the same 'switch' IP, but the port ID's are like: 109/1/3 + 104/1/43 for one host's pair of NIC's, and similar numbering for other ports (first number is 104,109,108, and 106 across 8 other NICs on 2 hosts.. third number ranges between 2 and 43). So I'm assuming that means they are indeed separate switches on a large chassis operating as one unit. – Crispin Bivans Dec 05 '13 at 20:39
  • Answer makes senses.. When I configured some virtual ports on the vDS, each vmkernel port got it's own IP and MAC Address per host.. There was no other way to configure it. So stands to reason, the LACP would have to be per host as well. Thanks. – Crispin Bivans Dec 10 '13 at 21:13