4

I have a cluster of 3 VMware ESXi6.0 hosts which are not licensed for VMware Distributed Switch. Currently, these hosts are configured with a static LAG between 2 ports on the server and a port each on a pair of Cisco Nexus 9372PX switches joined together with vPC. The servers connect via 10GbE over fibre.

My understanding of vPC is that it allows the switches to form aggregate links across multiple devices without the use of a stack. So, where I have configured a static LAG this seems to work fine.

My concern is that without a protocol such as LACP in place, should one of the Cisco switches go down, or should one of the server NICs fail, there is no way of renegotiating which ports can participate in the LAG so we would lose a theoretical 50% of traffic (based on Src/Dst IP Hash).

Is there a better way of configuring this link aggregation? On 10GbE, is there any merit in having an aggregated link (my gut feeling is no) and should I just let VMware "do it's thing" with regard to failover? What kind of configuration should I have in place on the switch?

I've done some research into this numerous times previously, and all roads seem to point to VMware Distributed Switch which, unfortunately, is prohibitively expensive for us.

1 Answers1

4

Good question.

There's no need to run static LAG in VMware. If you aren't in a position to use VMware Distributed Switches (vDs) and LACP, it's best to just use VMware's failover and set the NIC selection order in the Standard vSwitch and port group dialogs.

I don't know what your storage situation is, but for most port groups, setting all of the adapters to active is okay.

You can also override the order and pin a particular NIC as active and another as standby to control traffic flow.

Example VM networking port group: enter image description here

Example vmotion port group: enter image description here

ewwhite
  • 194,921
  • 91
  • 434
  • 799
  • Thanks for your feedback. On the other side of the servers we have a further 2x 10GbE NICs connected to the same Nexus switches in the same manner, which we use for our NFS connections back to a NetApp FAS. The NetApp is configured up for LACP already (and seems to run nicely with it). I'm guessing that the same configuration for our "LAN" side networking (i.e. breaking up the LAG and using VMware failover) will also be appropriate on the "SAN" side? – Daniel Arkley May 18 '16 at 18:59
  • 1
    Keep the LACP for the Netapp. For storage, break up the LAG and use VMware failover. It should look like my first image on the storage port group. – ewwhite May 18 '16 at 19:16