The Virtual Switch Eventually Had to Reach a Real Adapter
A Hyper-V virtual machine can have a virtual network adapter that behaves much like a physical network card from the guest operating system’s perspective. Behind that virtual device, however, network traffic eventually has to leave the host through actual hardware.
Hyper-V uses a virtual switch to connect virtual machines with each other and with external networks. When availability and additional network capacity were important, administrators could use more than one physical network adapter rather than depending on a single connection.
Windows Server 2016 introduced Switch Embedded Teaming, commonly called SET, which brought the teaming of those physical adapters directly into the Hyper-V virtual switch.
Virtual Networking Still Had a Physical Foundation
No matter how many virtual adapters existed inside the host, traffic leaving the server ultimately depended on the physical network interfaces connecting that machine to the network.
Several Physical Connections Could Behave Like a Larger Network Path
Network adapter teaming was not a new idea. Multiple physical network interfaces could be grouped so that traffic was not dependent on only one adapter.
This could improve availability because losing one physical connection did not necessarily mean losing the entire logical network connection. Depending on the configuration, traffic could also be distributed across multiple members of the team.
Traditional NIC teaming and the Hyper-V virtual switch, however, existed as distinct layers. SET changed where that teaming function could live.
The Team Could Move Inside the Virtual Switch
Instead of building a conventional adapter team first and then connecting the Hyper-V virtual switch to it, SET integrated the teaming functionality directly into the switch itself.
Hyper-V Could Bind Several Physical Adapters Directly
With Switch Embedded Teaming, multiple compatible physical network adapters could be attached to a Hyper-V virtual switch as members of the same SET team.
The virtual switch could then use those adapters collectively rather than depending on one physical interface beneath it.
This reduced the need for a separate traditional teaming layer in supported Hyper-V configurations and made adapter redundancy part of the virtual-switch architecture.
The Switch and the Team Were No Longer Separate Building Blocks
SET combined capabilities that previously could require a NIC team and a Hyper-V virtual switch to be configured as distinct networking components.
Another Physical Connection Could Remain Available
A server connected through only one network adapter has an obvious point of failure. The cable, adapter, switch port, or another part of that path can interrupt connectivity even when the server itself continues operating normally.
A SET team could contain multiple physical adapters. If one member became unavailable, the remaining member or members could continue providing connectivity according to the supported configuration.
The virtual machines above the switch therefore did not have to be tied to the health of one individual physical network interface.
Redundancy Could Exist Below the Virtual Machines
The guests could continue using their virtual network interfaces while the host networking layer handled the loss of an underlying physical team member.
A SET Team Could Include Several Physical Interfaces
Network redundancy is often illustrated with a simple pair of adapters, but SET was not limited to only two physical interfaces.
Windows Server 2016 allowed a SET team to contain as many as eight physical network adapters. This gave administrators flexibility when designing Hyper-V hosts with larger networking requirements.
The appropriate number of adapters depended on the server architecture and workload rather than on a requirement that every team look exactly the same.
The Team Could Grow Beyond a Simple Pair
Supporting multiple physical members allowed SET to participate in host designs where network availability and aggregate traffic requirements extended beyond two interfaces.
SET Was Designed Around Symmetrical Network Hardware
Combining network interfaces does not mean that arbitrary adapters should be mixed together. Predictable teaming behavior depends on the physical members having compatible characteristics.
SET required its member network adapters to be sufficiently alike, including requirements involving their speed and hardware characteristics.
This helped the virtual switch treat the physical members as a consistent group rather than attempting to distribute networking across mismatched hardware with substantially different capabilities.
More Adapters Did Not Mean Any Adapters
A successful SET design depended on compatible physical interfaces rather than simply placing whatever network cards happened to be available into the same team.
Physical Switches Could See Independent Connections
Some forms of network aggregation depend on coordination between the server and the physical network switch. That can require configuration on both sides of the connection.
SET was designed to operate in a switch-independent configuration. The physical network did not have to treat the host’s adapters as one specially coordinated switch-level aggregation group.
This could simplify the relationship between the Hyper-V host and the switches to which its physical network adapters were connected.
The Team Could Be a Host-Side Decision
The Hyper-V host could manage its SET membership without requiring the physical switches to build a corresponding dynamic teaming relationship.
The Physical Interfaces Did Not All Have to Sit Idle Behind One Active Link
Redundancy is useful even when a secondary connection spends most of its time waiting for a failure, but multiple network adapters can provide additional value when traffic is distributed across them.
SET supported load-distribution behavior that allowed networking activity to make use of the team’s physical members rather than treating every additional adapter solely as an unused standby.
The exact traffic path could vary with the networking configuration, but the larger idea was that the team could provide both resilience and useful network capacity.
Availability and Traffic Distribution Could Share the Same Hardware
The additional adapters could help preserve connectivity during failure while also participating in normal network operation when all team members were healthy.
High-Speed Storage Traffic Could Share the Virtual-Switch Architecture
Remote Direct Memory Access allows supported network adapters to move data with very low latency and reduced CPU involvement. That makes RDMA especially useful for demanding server and storage traffic.
Windows Server 2016 allowed RDMA-capable network adapters to participate in configurations using the Hyper-V virtual switch with Switch Embedded Teaming.
This was important because the same physical networking infrastructure could support virtualized networking while retaining capabilities needed for high-performance SMB and storage communication.
The Virtual Switch No Longer Had to Mean Giving Up RDMA
SET helped make it possible to combine Hyper-V networking and RDMA-oriented traffic within a converged host-network design instead of automatically dedicating entirely separate adapters to each purpose.
Fewer Physical Adapters Could Perform More Networking Jobs
Virtualization hosts can require connectivity for virtual machines, host management, storage, migration, and other infrastructure traffic. Building a separate set of physical network adapters for every category can quickly increase hardware and cabling requirements.
SET supported the broader move toward converged networking, where a smaller collection of capable physical adapters could carry several kinds of logically separated traffic.
The separation could exist through the virtual networking configuration even though the traffic ultimately shared the same underlying physical interfaces.
Logical Separation Did Not Always Require Physical Separation
Different networking purposes could be managed independently above the hardware while sharing a common set of high-performance physical connections underneath.
It Was Not Simply a New Name for Traditional NIC Teaming
Because both technologies combine physical network adapters, traditional NIC teaming and Switch Embedded Teaming can appear to solve exactly the same problem.
The important distinction was architectural. SET was integrated into the Hyper-V Virtual Switch and was designed for Hyper-V and software-defined networking scenarios rather than serving as a general replacement for every possible traditional teaming configuration.
That integration allowed the networking stack to coordinate capabilities that mattered specifically to modern virtualized hosts.
Similar Results Did Not Mean Identical Architecture
Both approaches could provide network redundancy, but SET placed the teaming function inside the Hyper-V virtual-switch path and therefore belonged to a different networking design.
Software Took Responsibility for More of the Host Network
Early virtualization networking could be understood largely as a way to connect virtual network adapters to physical networks. As server virtualization developed, the virtual switch became capable of doing considerably more.
Windows Server 2016 expanded software-defined networking capabilities around Hyper-V, and SET fit into that broader direction. Physical connectivity, traffic distribution, redundancy, and virtualized network policy increasingly met inside the software networking layer.
The host’s network architecture could therefore depend less on assembling separate hardware-oriented layers before the virtual switch was created.
The virtual switch could stop sitting on top of the network team because the network team could become part of the virtual switch itself.
Switch Embedded Teaming Brought Physical Redundancy Into the Hyper-V Switch
Switch Embedded Teaming changed the relationship between the Hyper-V Virtual Switch and the physical network adapters beneath it. Multiple compatible interfaces could become members of a team integrated directly into the virtual switch, providing redundancy and traffic distribution without requiring a separate traditional NIC team.
Support for multiple physical members, switch-independent operation, and RDMA-oriented configurations made SET particularly useful for virtualization hosts where several kinds of network traffic had to share resilient, high-performance connectivity.
The result was a more integrated host-network design. Instead of constructing the adapter team as a separate layer and placing the virtual switch above it, Windows Server could make the physical connections part of the virtual switch’s own networking architecture.
