A VM Normally Did Not Own the Physical Hardware It Used
Virtualization works largely by separating an operating system from the physical machine underneath it. A virtual machine sees processors, storage, network interfaces, and other devices presented through the virtualization environment rather than simply taking possession of the host’s hardware.
This abstraction makes virtual machines flexible. Hardware resources can be shared, controlled, moved, and managed without requiring every guest operating system to interact directly with each physical component.
There are situations, however, where that extra layer is not ideal. Some workloads benefit from reaching a physical device much more directly. Windows Server 2016 introduced Discrete Device Assignment, or DDA, to make that possible for compatible PCI Express hardware.
Virtualization Normally Hides the Physical Device
The guest generally works with hardware presented through Hyper-V rather than receiving unrestricted control of the actual PCIe component installed in the server.
The Hardware Could Be Removed From the Host and Given to a VM
Discrete Device Assignment allowed a compatible PCIe device to be dismounted from the host operating system and assigned directly to a particular virtual machine.
Once assigned, that physical device was no longer simply a host resource being indirectly shared with the guest. The VM gained direct access to the hardware.
This created a very different relationship between the virtual machine and the server’s physical components.
The Device Could Cross the Virtualization Boundary
Instead of Hyper-V presenting an abstract replacement for the hardware, DDA could place the actual PCIe device into the guest’s address space.
The Host and Guest Could Not Both Treat the Device as Their Own
Direct assignment did not mean that a physical device suddenly existed independently in two operating systems. The hardware had to be surrendered by the host before it could become available to the guest.
That exclusivity distinguished DDA from technologies designed to divide or share a device among multiple consumers.
The assigned VM effectively became the operating environment responsible for using that particular piece of hardware until the device was returned to the host.
Direct Meant Exclusive
A device assigned through DDA belonged to the selected virtual machine for the duration of the assignment rather than remaining simultaneously available as an ordinary host device.
Virtual Hardware Did Not Have to Pretend to Be the Physical Component
Because the actual PCIe device appeared inside the virtual machine, the guest could use the manufacturer’s appropriate device driver much as an operating system would when interacting with physical hardware.
This was useful for hardware whose capabilities mattered directly to applications running inside the VM.
The guest was no longer limited to a generalized virtual representation when a supported physical device had been deliberately assigned to it.
The Guest Could Recognize the Real Hardware
Once the device was mounted inside the VM, its normal manufacturer driver could be installed in the guest to operate the assigned component.
A Physical GPU Could Belong to a Virtual Machine
Graphics processors can perform far more than ordinary desktop display work. Engineering, visualization, computation, and other demanding applications may depend heavily on the capabilities of a particular GPU.
DDA provided a way for compatible graphics hardware to be assigned directly to a VM rather than requiring every GPU-dependent workload to run on the physical host.
The virtual machine could therefore remain an isolated operating environment while still gaining direct access to powerful physical graphics hardware.
A Virtual Workload Could Reach Physical Acceleration
Direct device assignment allowed certain hardware-intensive applications to remain inside a VM without necessarily giving up access to the underlying PCIe device they needed.
PCIe Hardware Was Not Limited to Graphics
The PCI Express bus connects many types of high-performance hardware. Storage devices using PCIe, including compatible NVMe devices, represented another important use for direct assignment.
Allowing a guest to communicate directly with such hardware could provide a different performance and control model from presenting storage through a conventional virtual disk.
The physical device itself could become part of the VM’s hardware environment rather than merely supplying capacity somewhere underneath a virtual storage layer.
The VM Could Receive the Device Rather Than Just Its Storage Space
Direct assignment changed the abstraction level by exposing compatible PCIe hardware itself instead of only presenting a virtual disk backed by physical storage.
The Data Path Could Bypass Part of Hyper-V
Every abstraction layer provides useful capabilities, but additional processing can also introduce overhead. A virtualized device normally requires Hyper-V to participate in the path between the guest and the physical hardware.
DDA was designed to bypass much of that virtualization path by allowing the guest to communicate directly with the assigned device.
For workloads where device performance was especially important, reducing the amount of virtualization between the application and hardware could be valuable.
Less Abstraction Could Mean Faster Hardware Access
The performance advantage came from allowing the VM to interact much more directly with the PCIe device instead of routing every operation through a fully virtualized hardware model.
Not Every Server Could Safely Hand Over a PCIe Device
Giving a guest direct control of physical hardware requires more than finding a PCIe slot and choosing a virtual machine. The server platform must be capable of safely controlling how device access is isolated.
Firmware and PCI Express capabilities therefore mattered. Windows needed sufficient control over the PCIe fabric, and the platform had to provide the hardware mechanisms required to isolate device traffic.
DDA was consequently tied to appropriate server-class hardware rather than being a feature that could be assumed to work on every computer capable of running Hyper-V.
Hyper-V Support Alone Was Not Enough
The physical server, firmware, PCIe topology, and selected device all had to provide the capabilities necessary for safe direct assignment.
A Directly Assigned Device Could Reach Beyond an Ordinary Virtual Peripheral
A virtual device exists within boundaries established by the virtualization platform. Passing an entire physical PCIe device into a guest changes that security relationship because the guest gains access to capabilities implemented by the device itself.
The platform therefore needed hardware-assisted isolation to prevent device activity from reaching memory or resources that did not belong to the assigned VM.
This is one reason DDA depended on specific platform capabilities rather than functioning as a simple software redirection technique.
Performance and Isolation Had to Exist Together
Direct access was useful only when the server could give the guest a short path to the hardware without sacrificing the boundaries protecting the rest of the system.
A VM Became Less Portable When Physical Hardware Belonged to It
One of virtualization’s greatest advantages is that a VM is not normally inseparable from a particular physical peripheral. That flexibility becomes harder to preserve when the guest has exclusive control of hardware installed in one server.
A directly assigned device cannot simply appear inside another host because the VM moves there. The corresponding physical component is still installed in the original machine.
As a result, some familiar Hyper-V mobility and state-management capabilities were restricted while a device was assigned through DDA.
Physical Attachment Reduced Virtual Mobility
Direct access traded some of the portability normally associated with a VM for a closer relationship with specific hardware installed in the host.
The VM Could Not Simply Carry Its PCIe Device to Another Host
Live migration normally allows a running virtual machine to move between compatible Hyper-V hosts with minimal interruption. That works because the VM’s virtualized resources can be reconstructed or made available at the destination.
A physical PCIe device assigned exclusively from one host creates a different problem. The hardware itself cannot migrate across the network with the VM.
Consequently, a VM using DDA could not use ordinary Hyper-V live migration while that device remained attached.
The Hardware Became an Anchor
Once a virtual machine depended directly on a particular physical device, the location of that device became part of where the VM could operate.
Direct Assignment Changed More Than Performance
The restrictions associated with DDA extended beyond live migration. Features that depend on Hyper-V having complete control over a VM’s resources could conflict with a guest directly owning physical hardware.
Save and restore functionality and Dynamic Memory were among the capabilities unavailable to a VM while using DDA.
This made direct assignment a deliberate architectural choice rather than a performance option that administrators could enable without considering its effect on the rest of the virtualization design.
A Faster Hardware Path Came With Conditions
DDA exchanged some virtualization flexibility for direct device access, making it important to decide whether the workload benefited enough from the physical hardware to justify those restrictions.
The Guest Could Remain Virtual Even When Its Hardware Was Real
Discrete Device Assignment illustrated an interesting change in the relationship between virtual and physical computing. The operating system, processors, memory environment, and other resources could remain part of a Hyper-V virtual machine while one selected component crossed that abstraction boundary.
The entire server did not have to be dedicated to the workload merely because an application needed direct access to specialized hardware.
Instead, Windows Server could keep the workload inside a VM while selectively allowing a compatible PCIe device to become effectively physical from the guest’s perspective.
The machine could remain virtual while one carefully selected piece of its hardware became very real.
Discrete Device Assignment Gave a VM Its Own Physical Hardware
Discrete Device Assignment introduced a different approach to hardware access in Hyper-V. Rather than always presenting a virtualized substitute, Windows Server could remove a compatible PCIe device from the host and assign it directly and exclusively to a virtual machine.
That arrangement could provide high-performance access to hardware such as compatible GPUs and PCIe storage devices while allowing the manufacturer’s driver to operate inside the guest. The closer connection to physical hardware also brought requirements and compromises, including suitable server hardware and restrictions on features such as live migration, save and restore, and Dynamic Memory.
DDA demonstrated that virtualization did not require every component to remain abstract. When a workload needed direct access badly enough, one physical device could cross the boundary and become part of the virtual machine itself.
