Starting with VxRail version 4.7.410, the customer can customize the teaming and failover policies for the VxRail networks. If your VxRail version is 7.0.010 or later, you can also customize the failover settings for the uplinks to an active-active configuration.
The decisions that you make on the teaming and failover policies are independent of the configuration on the adjacent top-of-rack switches. Be aware that load balancing has a dependency on physical switch settings, and requires link aggregation to be configured on the uplinks supporting the VxRail networks. Updating the teaming and failover policies with the top-of-rack switches will result in incremental improvements without also enabling link aggregation for load balancing purposes. Consideration should be given to coupling this effort with link aggregation to optimize bandwidth utilization.
During the initial implementation process, VxRail will apply a default teaming and failover policy for each VxRail network during the initial build operation.
- The default load-balancing policy is ‘Route based on originating virtual port’ for all VxRail network traffic. This policy directs network traffic up one uplink and does not attempt to share the workload between the uplinks.
- The default network failure detection setting is ‘link status only’. This setting should not be changed. VMware recommends having 3 or more physical NICs in the team for ‘beacon probing’ to work correctly, which is not supported with VxRail.
- The setting for ‘Notify switches’ is set to ‘Yes’. This instructs the virtual distributed switch to notify the adjacent physical switch of a failover.
- The setting for ‘Failback’ is set to ‘Yes’. This instructs a failed adapter to take over for the standby adapter once it is recovered and comes online again, if the uplinks are in an active-standby configuration.
- The failover order for the uplinks depends the VxRail network that is configured on the portgroup. VxRail assigns the uplinks to each portgroup depending on whether two ports or four ports were selected to support the VxRail cluster during initial implementation. For any given portgroup, one uplink is assigned ‘active’ mode, one uplink is assigned ‘standby’ mode, and any uplinks that are marked as ‘unused’ are in either ‘active’ or ‘standby’ mode in another portgroup.
Figure 44. Default VDS teaming and failover policy for vSAN network configured with 2 VxRail ports.
The following portgroup load-balancing policies are supported for VxRail clusters running version 4.7.410 or later:
- Route based on the originating virtual port ̶̶ After the virtual switch selects an uplink for a virtual machine or VMkernel adapter, it always forwards traffic through the same uplink. This option makes a simple selection based on the available physical uplinks. However, this policy does not attempt to load balance based on network traffic.
- Route based on source MAC hash ̶̶ The virtual switch selects an uplink for a virtual machine based on the virtual machine MAC address. While it requires more resources than using the originating virtual port, it has more flexibility in uplink selection. This policy does not attempt to load balance based on network traffic analysis.
- Use explicit failover order ̶ Always use the highest order uplink that passes failover detection criteria from the active adapters. No actual load balancing is performed with this option.
- Route based on physical NIC load ̶̶ The virtual switch monitors network traffic, and attempts to make adjustments on overloaded uplinks by moving traffic to another uplink. This option does use additional resources to track network traffic.
VxRail does not support the ‘Route based on IP Hash’ policy, as there is a dependency on the logical link setting of the physical port adapters on the switch. Link aggregation is required for this setting, which is introduced in version 7.0.130 of VxRail.
Starting with VxRail version 7.0.010, the ‘Failover Order’ setting on the teaming and failover policy on the VDS portgroups supporting VxRail networks can be changed. The default failover order for the uplinks each portgroup configured during VxRail initial build is described in Default failover order policy. For any portgroup configured during VxRail initial build to support VxRail network traffic, an uplink in ‘Standby’ mode can be moved into ‘Active’ mode to enable an ‘active/active’ configuration. This action can be performed after the VxRail cluster has completed the initial build operation.
Moving an uplink that is configured as ‘Unused’ for a portgroup supporting VxRail network traffic into either ‘Active’ mode or ‘Standby’ mode does not automatically activate the uplink and increase bandwidth for that portgroup. Bandwidth optimization depends the load-balancing settings on the upstream switches, and link aggregation is required to optimize network performance.
Figure 45. Sample failover order setting set to active/active