Attaching virtual GPU devices to guests

Attaching virtual GPU devices to guests

The virtual GPU feature in Nova allows a deployment to provide specific GPU types for instances using physical GPUs that can provide virtual devices.

For example, a single Intel GVT-g or a NVIDIA GRID vGPU physical Graphics Processing Unit (pGPU) can be virtualized as multiple virtual Graphics Processing Units (vGPUs) if the hypervisor supports the hardware driver and has the capability to create guests using those virtual devices.

This feature is highly dependent on the hypervisor, its version and the physical devices present on the host.

Important

As of the Queens release, there is no upstream continuous integration testing with a hardware environment that has virtual GPUs and therefore this feature is considered experimental.

Hypervisor-specific caveats are mentioned in the Caveats section.

To enable virtual GPUs, follow the steps below:

  1. Enable GPU types (Compute)
  2. Configure a flavor (Controller)

Enable GPU types (Compute)

  1. Specify which specific GPU type(s) the instances would get.

    Edit devices.enabled_vgpu_types:

    [devices]
    enabled_vgpu_types = nvidia-35
    

    Note

    As of the Queens release, Nova only supports a single type. If more than one vGPU type is specified (as a comma-separated list), only the first one will be used.

    To know which specific type(s) to mention, please refer to How to discover a GPU type.

  2. Restart the nova-compute service.

Configure a flavor (Controller)

Configure a flavor to request one virtual GPU:

$ openstack flavor set vgpu_1 --property "resources:VGPU=1"

Note

As of the Queens release, all hypervisors that support virtual GPUs only accept a single virtual GPU per instance.

The enabled vGPU types on the compute hosts are not exposed to API users. Flavors configured for vGPU support can be tied to host aggregates as a means to properly schedule those flavors onto the compute hosts that support them. See the host aggregates guide for more information.

Create instances with virtual GPU devices

The nova-scheduler selects a destination host that has vGPU devices available by calling the Placement API for a specific VGPU resource class provided by compute nodes.

$ openstack server create --flavor vgpu_1 --image cirros-0.3.5-x86_64-uec --wait test-vgpu

Note

As of the Queens release, only the FilterScheduler scheduler driver uses the Placement API.

How to discover a GPU type

Depending on your hypervisor:

  • For libvirt, virtual GPUs are seen as mediated devices. Physical PCI devices (the graphic card here) supporting virtual GPUs propose mediated device (mdev) types. Since mediated devices are supported by the Linux kernel through sysfs files, you can see the required properties as follows:

    $ ls /sys/class/mdev_bus/*/mdev_supported_types
    /sys/class/mdev_bus/0000:84:00.0/mdev_supported_types:
    nvidia-35  nvidia-36  nvidia-37  nvidia-38  nvidia-39  nvidia-40  nvidia-41  nvidia-42  nvidia-43  nvidia-44  nvidia-45
    
    /sys/class/mdev_bus/0000:85:00.0/mdev_supported_types:
    nvidia-35  nvidia-36  nvidia-37  nvidia-38  nvidia-39  nvidia-40  nvidia-41  nvidia-42  nvidia-43  nvidia-44  nvidia-45
    
    /sys/class/mdev_bus/0000:86:00.0/mdev_supported_types:
    nvidia-35  nvidia-36  nvidia-37  nvidia-38  nvidia-39  nvidia-40  nvidia-41  nvidia-42  nvidia-43  nvidia-44  nvidia-45
    
    /sys/class/mdev_bus/0000:87:00.0/mdev_supported_types:
    nvidia-35  nvidia-36  nvidia-37  nvidia-38  nvidia-39  nvidia-40  nvidia-41  nvidia-42  nvidia-43  nvidia-44  nvidia-45
    
  • For XenServer, virtual GPU types are created by XenServer at startup depending on the available hardware and config files present in dom0. You can run the command of xe vgpu-type-list from dom0 to get the available vGPU types. The value for the field of model-name ( RO): is the vGPU type's name which can be used to set the nova config option [devices]/enabled_vgpu_types. See the following example:

    [root@trailblazer-2 ~]# xe vgpu-type-list
    uuid ( RO)              : 78d2d963-41d6-4130-8842-aedbc559709f
           vendor-name ( RO): NVIDIA Corporation
            model-name ( RO): GRID M60-8Q
             max-heads ( RO): 4
        max-resolution ( RO): 4096x2160
    
    
    uuid ( RO)              : a1bb1692-8ce3-4577-a611-6b4b8f35a5c9
           vendor-name ( RO): NVIDIA Corporation
            model-name ( RO): GRID M60-0Q
             max-heads ( RO): 2
        max-resolution ( RO): 2560x1600
    
    
    uuid ( RO)              : 69d03200-49eb-4002-b661-824aec4fd26f
           vendor-name ( RO): NVIDIA Corporation
            model-name ( RO): GRID M60-2A
             max-heads ( RO): 1
        max-resolution ( RO): 1280x1024
    
    
    uuid ( RO)              : c58b1007-8b47-4336-95aa-981a5634d03d
           vendor-name ( RO): NVIDIA Corporation
            model-name ( RO): GRID M60-4Q
             max-heads ( RO): 4
        max-resolution ( RO): 4096x2160
    
    
    uuid ( RO)              : 292a2b20-887f-4a13-b310-98a75c53b61f
           vendor-name ( RO): NVIDIA Corporation
            model-name ( RO): GRID M60-2Q
             max-heads ( RO): 4
        max-resolution ( RO): 4096x2160
    
    
    uuid ( RO)              : d377db6b-a068-4a98-92a8-f94bd8d6cc5d
           vendor-name ( RO): NVIDIA Corporation
            model-name ( RO): GRID M60-0B
             max-heads ( RO): 2
        max-resolution ( RO): 2560x1600
    
    ...
    

Caveats

Note

This information is correct as of the 17.0.0 Queens release. Where improvements have been made or issues fixed, they are noted per item.

For libvirt:

  • Suspending a guest that has vGPUs doesn't yet work because of a libvirt limitation (it can't hot-unplug mediated devices from a guest). Workarounds using other instance actions (like snapshotting the instance or shelving it) are recommended until libvirt gains mdev hot-unplug support. If a user attempts to suspend the instance, the libvirt driver will raise an exception that will cause the instance to be set back to ACTIVE. The suspend action in the os-instance-actions API will have an Error state.

  • Resizing an instance with a new flavor that has vGPU resources doesn't allocate those vGPUs to the instance (the instance is created without vGPU resources). The proposed workaround is to rebuild the instance after resizing it. The rebuild operation allocates vGPUS to the instance.

  • Cold migrating an instance to another host will have the same problem as resize. If you want to migrate an instance, make sure to rebuild it after the migration.

  • Rescue images do not use vGPUs. An instance being rescued does not keep its vGPUs during rescue. During that time, another instance can receive those vGPUs. This is a known issue. The recommended workaround is to rebuild an instance immediately after rescue. However, rebuilding the rescued instance only helps if there are other free vGPUs on the host.

    Note

    This has been resolved in the Rocky release [1].

For XenServer:

  • Suspend and live migration with vGPUs attached depends on support from the underlying XenServer version. Please see XenServer release notes for up to date information on when a hypervisor supporting live migration and suspend/resume with vGPUs is available. If a suspend or live migrate operation is attempted with a XenServer version that does not support that operation, an internal exception will occur that will cause nova setting the instance to be in ERROR status. You can use the command of openstack server set --state active <server> to set it back to ACTIVE.
  • Resizing an instance with a new flavor that has vGPU resources doesn't allocate those vGPUs to the instance (the instance is created without vGPU resources). The proposed workaround is to rebuild the instance after resizing it. The rebuild operation allocates vGPUS to the instance.
  • Cold migrating an instance to another host will have the same problem as resize. If you want to migrate an instance, make sure to rebuild it after the migration.
[1]https://bugs.launchpad.net/nova/+bug/1762688
Creative Commons Attribution 3.0 License

Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.