VMware Solutions Discussions

Layering vif's where lacp member vif's are of unequal size


Hi, I have a dual-controller 3140a in a QA environment with only 2 on-board GigE ports each.  The current config has:

Single vif per controller

- e0a active

- e0b standby

- each nic port is attached to a separate Cisco 6509-E switch.  There is a 4-gig port channel between the switches but they do not have the cross-stacking VSS module.

On each side I am going to add a 4-port GigE PCIe card and then implement:

- 2 lacp vif's - one connected to each 6509 switch - vif_lacp0, vif_lacp1

- 1 single-mode vif layered above the 2 lacp vif's.

So the config on Controller1 would look like:

vif create lacp -b ip vif_lacp0 <nics>

vif create lacp -b ip vif_lacp1 <nics>

vif create single vif_single0 vif_lacp0 vif_lacp1

vif favor vif_lacp0

* matching config with appropriate naming changes will be put on Controller2

- Controller 1 is the primary controller for Oracle DB connections (2 2-node RAC clusters)

- Controller 2 is the primary controller for a VMware vSphere cluster (8 hosts)

What I'm trying to determine and decide is whether the <nics>  belonging to the lacp vif's have to be equal in number?  If not I was considering making:

vif_lacp0 (active) use all 4-ports from the add-in card

vif_lacp1 (standby) use the 2-ports from the on-board card


- additional bandwidth with higher single link-loss failure on active vif's, which should be up and active almost 100% of the time

- avoid any issues if differences in chipsets b/w onboard nics and expansion card nics


- Lower bandwidth available during outages, although it would take full loss of primary 6500 to be in that situation

Anyone tried this before with good or bad results?





Yes this does work and is supported.  Another example we have implemented in a similar manner is a single 10GbE NIC as primary then a 4x 1GbE LACP secondary.  Then the single mode vif contains the single interface and the second level vif.