Virtual machine workloads share not only the compute and storage resources on an ESXi server, but the physical network interfaces as well. There are physical limits on how well these network interfaces can serve the bandwidth needs of the virtual machines using them. More importantly, not every virtual machine has the same network workload characteristics. Hence, it becomes extremely critical for the virtual switches to have network load distribution, load balancing, and failover capabilities. 

Let's begin by comparing the teaming and failover capabilities of both the switch types:

Route based on the originating virtual port, With this mechanism, every virtual port to which a vNIC connects is associated with an uplink (vmnic) in a round-robin fashion. This means that, if there were only two physical uplinks—vmnic-A/dvUplink-A and vmnic-B/dvuplink-B, and four virtual ports—Port1, Port2, Port3, and Port4, then Port1, and Port3 will be associated with vmnic-A/dvUplink-A and Port2, and Port4 will be associated with vmnic-B/dvUplink-B:

 

Once associated, the uplinks will be used for traffic unless there is an uplink failure, a VMotion, or a power-cycle of the VM. This is the default load balancing mechanism chosen for both vSwitch and dvSwitch:

• Route based on source MAC Hash: This is a deprecated method of load balancing. An uplink is chosen based on the source MAC address of the frames that enter the virtual switch.
• Route based on IP Hash: This method uses the combined hash value of the source and destination IP addresses to choose an uplink. The physical NICs must be in the same link aggregation group. If the physical NICs are cabled to different switches then those switches should be stacked. The route based on the IP hash is particularly useful if there are multiple source and destination IP addresses to calculate hashes from. For instance, if you have a web server virtual machine with clients from a different subnet then such a combination is an ideal candidate for the route based on the IP hash algorithm.
• Load Based Teaming (LBT): This load balancing mechanism is only available on a dvSwitch. Unlike the other methods, this offers true load balancing. The other methods only offer the distribution of the physical adapter assignments based on their algorithms. With LBT, the initial assignment is done in a round-robin fashion, but from then on, the physical adapters are monitored for load saturation. If any of the physical adapters hit a saturation threshold of 75% that persists for over 30 seconds, then some of the traffic is relocated to another unsaturated physical adapter.
• Use explicit failover: This is not a load balancing or distribution method; instead, it uses a pre-defined failover order to use the active, available physical adapters in the event of a failure. When you set load balancing to use explicit failover order, all the traffic is traversed through a single physical adapter at any point in time. If there is more than one active physical adapter, then it will choose the adapter that has been up and active for the longest time. For instance, if vmnic1, vmnic2, and vmnic3 are the three active uplinks, and have an uptime of 48 hours, 32 hours, and 5 minutes respectively, then vmnic1 with 48 hours of uptime is chosen. The logic behind such a choice is to select the most stable among the available adapters.