I am not the authority on Best practice for harvester but I have done a ton of infrastructure work. Depending on the criticality of the system, I split my nics as follows: Mgmt - 1gb (if I have two nics and management switches I bond them) • Management traffic for managing the system only backend - 10GB or above • VM Migration • Storage - I split this out to separate physical nics if I can or give priority over all other traffic. My Migrations can wait for storage traffic frontend - 10GB or above • VM to VM traffic • Public NIC traffic • Tenant Traffic If you only have 2 sets of NICs (i.e. your 2x2) then I would split management and VM traffic The other way of thinking is that the physical machines (node) are commodity so single connect things unless you need the bandwidth and when a nic fails or you need to do maintenance fail it over and move or restart the machines on another nodes. Just my two cents.