Chapter 8. Configuring virtual GPUs for instances

To support GPU-based rendering on your instances, you can define and manage virtual GPU (vGPU) resources according to your available physical GPU devices and your hypervisor type. You can use this configuration to divide the rendering workloads between all your physical GPU devices more effectively, and to have more control over scheduling your vGPU-enabled instances.

To enable vGPU in OpenStack Compute, create flavors that your cloud users can use to create Red Hat Enterprise Linux (RHEL) instances with vGPU devices. Each instance can then support GPU workloads with virtual GPU devices that correspond to the physical GPU devices.

The OpenStack Compute service tracks the number of vGPU devices that are available for each GPU profile you define on each host. The Compute service schedules instances to these hosts based on the flavor, attaches the devices, and monitors usage on an ongoing basis. When an instance is deleted, the Compute service adds the vGPU devices back to the available pool.

8.1. Supported configurations and limitations

Supported GPU cards

For a list of supported NVIDIA GPU cards, see Virtual GPU Software Supported Products on the NVIDIA website.

Limitations when using vGPU devices

  • You can enable only one vGPU type on each Compute node.
  • Each instance can use only one vGPU resource.
  • Live migration of vGPU between hosts is not supported.
  • Suspend operations on a vGPU-enabled instance is not supported due to a libvirt limitation. Instead, you can snapshot or shelve the instance.
  • Resize and cold migration operations on an instance with a vGPU flavor does not automatically re-allocate the vGPU resources to the instance. After you resize or migrate the instance, you must rebuild it manually to re-allocate the vGPU resources.
  • By default, vGPU types on Compute hosts are not exposed to API users. To grant access, add the hosts to a host aggregate. For more information, see Section 4.4, “Manage Host Aggregates”.
  • If you use NVIDIA accelerator hardware, you must comply with the NVIDIA licensing requirements. For example, NVIDIA vGPU GRID requires a licensing server. For more information about the NVIDIA licensing requirements, see NVIDIA License Server Release Notes on the NVIDIA website.

8.2. Configuring vGPU on the Compute nodes

To enable your cloud users to create instances that use a virtual GPU (vGPU), you must configure the Compute nodes that have the physical GPUs:

  1. Build a custom GPU-enabled overcloud image.
  2. Prepare the GPU role, profile, and flavor for designating Compute nodes for vGPU.
  3. Configure the Compute node for vGPU.
  4. Deploy the overcloud.
Note

To use an NVIDIA GRID vGPU, you must comply with the NVIDIA GRID licensing requirements and you must have the URL of your self-hosted license server. For more information, see the NVIDIA License Server Release Notes web page.

8.2.1. Building a custom GPU overcloud image

Perform the following steps on the director node to install the NVIDIA GRID host driver on an overcloud Compute image and upload the image to the OpenStack Image Service (glance).

Procedure

  1. Copy the overcloud image and add the gpu suffix to the copied image.

    $ cp overcloud-full.qcow2 overcloud-full-gpu.qcow2
  2. Install an ISO image generator tool from YUM.

    $ sudo yum install genisoimage -y
  3. Download the NVIDIA GRID host driver RPM package that corresponds to your GPU device from the NVIDIA website. To determine which driver you need, see the NVIDIA Driver Downloads Portal.

    Note

    You must be a registered NVIDIA customer to download the drivers from the portal.

  4. Create an ISO image from the driver RPM package and save the image in the nvidia-host directory.

    $ genisoimage -o nvidia-host.iso -R -J -V NVIDIA nvidia-host/
    I: -input-charset not specified, using utf-8 (detected in locale settings)
      9.06% done, estimate finish Wed Oct 31 11:24:46 2018
     18.08% done, estimate finish Wed Oct 31 11:24:46 2018
     27.14% done, estimate finish Wed Oct 31 11:24:46 2018
     36.17% done, estimate finish Wed Oct 31 11:24:46 2018
     45.22% done, estimate finish Wed Oct 31 11:24:46 2018
     54.25% done, estimate finish Wed Oct 31 11:24:46 2018
     63.31% done, estimate finish Wed Oct 31 11:24:46 2018
     72.34% done, estimate finish Wed Oct 31 11:24:46 2018
     81.39% done, estimate finish Wed Oct 31 11:24:46 2018
     90.42% done, estimate finish Wed Oct 31 11:24:46 2018
     99.48% done, estimate finish Wed Oct 31 11:24:46 2018
    Total translation table size: 0
    Total rockridge attributes bytes: 358
    Total directory bytes: 0
    Path table size(bytes): 10
    Max brk space used 0
    55297 extents written (108 MB)
  5. Create a driver installation script for your Compute nodes. This script installs the NVIDIA GRID host driver on each Compute node that you run it on. The following example creates a script named install_nvidia.sh:

    #/bin/bash
    
    # NVIDIA GRID package
    mkdir /tmp/mount
    mount LABEL=NVIDIA /tmp/mount
    rpm -ivh /tmp/mount/NVIDIA-vGPU-rhel-8.1-430.27.x86_64.rpm
  6. Customize the overcloud image by attaching the ISO image that you generated in Step 4, and running the driver installation script that you created in Step 5:

    $ virt-customize --attach nvidia-packages.iso -a overcloud-full-gpu.qcow2  -v --run install_nvidia.sh
    [   0.0] Examining the guest ...
    libguestfs: launch: program=virt-customize
    libguestfs: launch: version=1.36.10rhel=8,release=6.el8_5.2,libvirt
    libguestfs: launch: backend registered: unix
    libguestfs: launch: backend registered: uml
    libguestfs: launch: backend registered: libvirt
  7. Relabel the customized image with SELinux:

    $ virt-customize -a overcloud-full-gpu.qcow2 --selinux-relabel
    [   0.0] Examining the guest ...
    [   2.2] Setting a random seed
    [   2.2] SELinux relabelling
    [  27.4] Finishing off
  8. Prepare the custom image files for upload to the OpenStack Image Service:

    $ mkdir /var/image/x86_64/image
    $ guestmount -a overcloud-full-gpu.qcow2 -i --ro image
    $ cp image/boot/vmlinuz-3.10.0-862.14.4.el8.x86_64 ./overcloud-full-gpu.vmlinuz
    $ cp image/boot/initramfs-3.10.0-862.14.4.el8.x86_64.img ./overcloud-full-gpu.initrd
  9. From the undercloud, upload the custom image to the OpenStack Image Service:

    (undercloud) $ openstack overcloud image upload --update-existing --os-image-name overcloud-full-gpu.qcow2

8.2.2. Designating Compute nodes for vGPU

To designate Compute nodes for vGPU workloads, you must create a new role file to configure the vGPU role, and configure a new flavor to use to tag the GPU-enabled Compute nodes.

Procedure

  1. To create the new ComputeGPU role file, copy the file /usr/share/openstack-tripleo-heat-templates/roles/Compute.yaml to /usr/share/openstack-tripleo-heat-templates/roles/ComputeGPU.yaml and edit the following file sections:

    Table 8.1. ComputeGPU role file edits

    Section/ParameterCurrent valueNew value

    Role comment

    Role: Compute

    Role: ComputeGpu

    Role name

    name: Compute

    name: ComputeGpu

    description

    Basic Compute Node role

    GPU Compute Node role

    ImageDefault

    overcloud-full

    overcloud-full-gpu

    HostnameFormatDefault

    -compute-

    -computegpu-

    deprecated_nic_config_name

    compute.yaml

    compute-gpu.yaml

  2. Generate a new roles data file named gpu_roles_data.yaml that includes the Controller, Compute, and ComputeGpu roles.

    (undercloud) [stack@director templates]$ openstack overcloud roles generate -o /home/stack/templates/gpu_roles_data.yaml Controller Compute ComputeGpu

    The following example shows the ComputeGpu role details:

    #####################################################################
    # Role: ComputeGpu                                                  #
    #####################################################################
    - name: ComputeGpu
      description: |
        GPU Compute Node role
      CountDefault: 1
      ImageDefault: overcloud-full-gpu
      networks:
        - InternalApi
        - Tenant
        - Storage
      HostnameFormatDefault: '%stackname%-computegpu-%index%'
      RoleParametersDefault:
        TunedProfileName: "virtual-host"
      # Deprecated & backward-compatible values (FIXME: Make parameters consistent)
      # Set uses_deprecated_params to True if any deprecated params are used.
      uses_deprecated_params: True
      deprecated_param_image: 'NovaImage'
      deprecated_param_extraconfig: 'NovaComputeExtraConfig'
      deprecated_param_metadata: 'NovaComputeServerMetadata'
      deprecated_param_scheduler_hints: 'NovaComputeSchedulerHints'
      deprecated_param_ips: 'NovaComputeIPs'
      deprecated_server_resource_name: 'NovaCompute'
      deprecated_nic_config_name: 'compute-gpu.yaml'
      ServicesDefault:
        - OS::TripleO::Services::Aide
        - OS::TripleO::Services::AuditD
        - OS::TripleO::Services::BootParams
        - OS::TripleO::Services::CACerts
        - OS::TripleO::Services::CephClient
        - OS::TripleO::Services::CephExternal
        - OS::TripleO::Services::CertmongerUser
        - OS::TripleO::Services::Collectd
        - OS::TripleO::Services::ComputeCeilometerAgent
        - OS::TripleO::Services::ComputeNeutronCorePlugin
        - OS::TripleO::Services::ComputeNeutronL3Agent
        - OS::TripleO::Services::ComputeNeutronMetadataAgent
        - OS::TripleO::Services::ComputeNeutronOvsAgent
        - OS::TripleO::Services::Docker
        - OS::TripleO::Services::Fluentd
        - OS::TripleO::Services::IpaClient
        - OS::TripleO::Services::Ipsec
        - OS::TripleO::Services::Iscsid
        - OS::TripleO::Services::Kernel
        - OS::TripleO::Services::LoginDefs
        - OS::TripleO::Services::MetricsQdr
        - OS::TripleO::Services::MySQLClient
        - OS::TripleO::Services::NeutronBgpVpnBagpipe
        - OS::TripleO::Services::NeutronLinuxbridgeAgent
        - OS::TripleO::Services::NeutronVppAgent
        - OS::TripleO::Services::NovaCompute
        - OS::TripleO::Services::NovaLibvirt
        - OS::TripleO::Services::NovaLibvirtGuests
        - OS::TripleO::Services::NovaMigrationTarget
        - OS::TripleO::Services::ContainersLogrotateCrond
        - OS::TripleO::Services::OpenDaylightOvs
        - OS::TripleO::Services::Podman
        - OS::TripleO::Services::Rhsm
        - OS::TripleO::Services::RsyslogSidecar
        - OS::TripleO::Services::Securetty
        - OS::TripleO::Services::SensuClient
        - OS::TripleO::Services::Snmp
        - OS::TripleO::Services::Sshd
        - OS::TripleO::Services::Timesync
        - OS::TripleO::Services::Timezone
        - OS::TripleO::Services::TripleoFirewall
        - OS::TripleO::Services::TripleoPackages
        - OS::TripleO::Services::Tuned
        - OS::TripleO::Services::Vpp
        - OS::TripleO::Services::OVNController
        - OS::TripleO::Services::OVNMetadataAgent
  3. Register the node for the overcloud. For more information, see Registering nodes for the overcloud in the Director Installation and Usage guide.
  4. Inspect the node hardware. For more information, see Inspecting the hardware of nodes in the Director Installation and Usage guide.
  5. Create the compute-vgpu-nvidia flavor to use to tag nodes that you want to designate for vGPU workloads:

    (undercloud) [stack@director templates]$ openstack flavor create --id auto --ram 6144 --disk 40 --vcpus 4 compute-vgpu-nvidia
    +----------------------------+--------------------------------------+
    | Field                      | Value                                |
    +----------------------------+--------------------------------------+
    | OS-FLV-DISABLED:disabled   | False                                |
    | OS-FLV-EXT-DATA:ephemeral  | 0                                    |
    | disk                       | 40                                   |
    | id                         | 9cb47954-be00-47c6-a57f-44db35be3e69 |
    | name                       | compute-vgpu-nvidia                  |
    | os-flavor-access:is_public | True                                 |
    | properties                 |                                      |
    | ram                        | 6144                                 |
    | rxtx_factor                | 1.0                                  |
    | swap                       |                                      |
    | vcpus                      | 4                                    |
    +----------------------------+--------------------------------------+
  6. Tag each node that you want to designate for GPU workloads with the compute-vgpu-nvidia profile.

    (undercloud) [stack@director templates]$ openstack baremetal node set --property capabilities='profile:compute-vgpu-nvidia,boot_option:local' <node>

    Replace <node> with the ID of the baremetal node.

8.2.3. Configuring the Compute node for vGPU and deploying the overcloud

You need to retrieve and assign the vGPU type that corresponds to the physical GPU device in your environment, and prepare the environment files to configure the Compute node for vGPU.

Procedure

  1. Install Red Hat Enterprise Linux and the NVIDIA GRID driver on a temporary Compute node and launch the node. For more information about installing the NVIDIA GRID driver, see Section 8.2.1, “Building a custom GPU overcloud image”.
  2. On the Compute node, locate the vGPU type of the physical GPU device that you want to enable. For libvirt, virtual GPUs are mediated devices, or mdev type devices. To discover the supported mdev devices, enter the following command:

    [root@overcloud-computegpu-0 ~]# ls /sys/class/mdev_bus/0000\:06\:00.0/mdev_supported_types/
    nvidia-11  nvidia-12  nvidia-13  nvidia-14  nvidia-15  nvidia-16  nvidia-17  nvidia-18  nvidia-19  nvidia-20  nvidia-21  nvidia-210  nvidia-22
    
    [root@overcloud-computegpu-0 ~]# cat /sys/class/mdev_bus/0000\:06\:00.0/mdev_supported_types/nvidia-18/description
    num_heads=4, frl_config=60, framebuffer=2048M, max_resolution=4096x2160, max_instance=4
  3. Add the compute-gpu.yaml file to the network-environment.yaml file:

    resource_registry:
      OS::TripleO::Compute::Net::SoftwareConfig: /home/stack/templates/nic-configs/compute.yaml
      OS::TripleO::ComputeGpu::Net::SoftwareConfig: /home/stack/templates/nic-configs/compute-gpu.yaml
      OS::TripleO::Controller::Net::SoftwareConfig: /home/stack/templates/nic-configs/controller.yaml
      #OS::TripleO::AllNodes::Validation: OS::Heat::None
  4. Add the following parameters to the node-info.yaml file to specify the number of GPU Compute nodes, and the flavor to use for the GPU-designated Compute nodes:

    parameter_defaults:
      OvercloudControllerFlavor: control
      OvercloudComputeFlavor: compute
      OvercloudComputeGpuFlavor: compute-vgpu-nvidia
      ControllerCount: 1
      ComputeCount: 0
      ComputeGpuCount: 1
  5. Create a gpu.yaml file to specify the vGPU type of your GPU device:

    parameter_defaults:
      ComputeGpuExtraConfig:
        nova::compute::vgpu::enabled_vgpu_types:
          - nvidia-18
    Note

    Each physical GPU supports only one virtual GPU type. If you specify multiple vGPU types in this property, only the first type is used.

  6. Deploy the overcloud, adding your new role and environment files to the stack along with your other environment files:

    (undercloud) $ openstack overcloud deploy --templates \
      -r /home/stack/templates/nvidia/gpu_roles_data.yaml
      -e /home/stack/templates/node-info.yaml
      -e /home/stack/templates/network-environment.yaml
      -e [your environment files]
      -e /home/stack/templates/gpu.yaml

8.3. Creating the vGPU image and flavor

To enable your cloud users to create instances that use a virtual GPU (vGPU), you can define a custom vGPU-enabled image, and you can create a vGPU flavor.

8.3.1. Creating a custom GPU instance image

After you deploy the overcloud with GPU-enabled Compute nodes, you can create a custom vGPU-enabled instance image with the NVIDIA GRID guest driver and license file.

Procedure

  1. Create an instance with the hardware and software profile that your vGPU instances require:

    (overcloud) [stack@director ~]$ openstack server create --flavor <flavor> --image <image> temp_vgpu_instance
    • Replace <flavor> with the name or ID of the flavor that has the hardware profile that your vGPU instances require. For information on default flavors, see Manage flavors.
    • Replace <image> with the name or ID of the image that has the software profile that your vGPU instances require. For information on downloading RHEL cloud images, see Image service.
  2. Log in to the instance as a cloud-user. For more information, see Log in to an Instance.
  3. Create the gridd.conf NVIDIA GRID license file on the instance, following the NVIDIA guidance: Licensing an NVIDIA vGPU on Linux by Using a Configuration File.
  4. Install the GPU driver on the instance. For more information about installing an NVIDIA driver, see Installing the NVIDIA vGPU Software Graphics Driver on Linux.

    Note

    Use the hw_video_model image property to define the GPU driver type. You can choose none if you want to disable the emulated GPUs for your vGPU instances. For more information about supported drivers, see Appendix A, Image Configuration Parameters.

  5. Create an image snapshot of the instance:

    (overcloud) [stack@director ~]$ openstack server image create --name vgpu_image temp_vgpu_instance
  6. Optional: Delete the instance.

8.3.2. Creating a vGPU flavor for instances

After you deploy the overcloud with GPU-enabled Compute nodes, you can create a custom flavor that your cloud users can use to launch instances for GPU workloads.

Procedure

  1. Create an NVIDIA GPU flavor. For example:

    (overcloud) [stack@virtlab-director2 ~]$ openstack flavor create --vcpus 6 --ram 8192 --disk 100 m1.small-gpu
    +----------------------------+--------------------------------------+
    | Field                      | Value                                |
    +----------------------------+--------------------------------------+
    | OS-FLV-DISABLED:disabled   | False                                |
    | OS-FLV-EXT-DATA:ephemeral  | 0                                    |
    | disk                       | 100                                  |
    | id                         | a27b14dd-c42d-4084-9b6a-225555876f68 |
    | name                       | m1.small-gpu                         |
    | os-flavor-access:is_public | True                                 |
    | properties                 |                                      |
    | ram                        | 8192                                 |
    | rxtx_factor                | 1.0                                  |
    | swap                       |                                      |
    | vcpus                      | 6                                    |
    +----------------------------+--------------------------------------+
  2. Assign a vGPU resource to the flavor that you created. You can assign only one vGPU for each instance.

    (overcloud) [stack@virtlab-director2 ~]$ openstack flavor set m1.small-gpu --property "resources:VGPU=1"
    
    (overcloud) [stack@virtlab-director2 ~]$ openstack flavor show m1.small-gpu
    +----------------------------+--------------------------------------+
    | Field                      | Value                                |
    +----------------------------+--------------------------------------+
    | OS-FLV-DISABLED:disabled   | False                                |
    | OS-FLV-EXT-DATA:ephemeral  | 0                                    |
    | access_project_ids         | None                                 |
    | disk                       | 100                                  |
    | id                         | a27b14dd-c42d-4084-9b6a-225555876f68 |
    | name                       | m1.small-gpu                         |
    | os-flavor-access:is_public | True                                 |
    | properties                 | resources:VGPU='1'                   |
    | ram                        | 8192                                 |
    | rxtx_factor                | 1.0                                  |
    | swap                       |                                      |
    | vcpus                      | 6                                    |
    +----------------------------+--------------------------------------+

8.3.3. Launching a vGPU instance

You can create a GPU-enabled instance for GPU workloads.

Procedure

  1. Create an instance using a GPU flavor and image. For example:

    (overcloud) [stack@virtlab-director2 ~]$ openstack server create --flavor m1.small-gpu --image vgpu_image --security-group web --nic net-id=internal0 --key-name lambda vgpu-instance
  2. Log in to the instance as a cloud-user. For more information, see Log in to an Instance.
  3. To verify that the GPU is accessible from the instance, run the following command from the instance:

    $ lspci -nn | grep <gpu_name>

8.4. Enabling PCI passthrough for a GPU device

You can use PCI passthrough to attach a physical PCI device, such as a graphics card, to an instance. If you use PCI passthrough for a device, the instance reserves exclusive access to the device for performing tasks, and the device is not available to the host.

Prerequisites

Procedure

  1. To determine the vendor ID and product ID for each passthrough device type, run the following command on the physical server that has the PCI cards:

    # lspci -nn | grep -i <gpu_name>

    For example, to determine the vendor and product ID for an NVIDIA GPU, run the following command:

    # lspci -nn | grep -i nvidia
    3b:00.0 3D controller [0302]: NVIDIA Corporation TU104GL [Tesla T4] [10de:1eb8] (rev a1)
    d8:00.0 3D controller [0302]: NVIDIA Corporation TU104GL [Tesla T4] [10de:1db4] (rev a1)
  2. To configure the Controller node on the overcloud for PCI passthrough, create an environment file, for example, pci_passthru_controller.yaml.
  3. Add PciPassthroughFilter to the NovaSchedulerDefaultFilters parameter in pci_passthru_controller.yaml:

    parameter_defaults:
      NovaSchedulerDefaultFilters: ['RetryFilter','AvailabilityZoneFilter','ComputeFilter','ComputeCapabilitiesFilter','ImagePropertiesFilter','ServerGroupAntiAffinityFilter','ServerGroupAffinityFilter','PciPassthroughFilter','NUMATopologyFilter']
  4. To specify the PCI alias for the devices on the Controller node, add the following to pci_passthru_controller.yaml:

    ControllerExtraConfig:
        nova::pci::aliases:
          -  name: "t4"
             product_id: "1eb8"
             vendor_id: "10de"
          -  name: "v100"
             product_id: "1db4"
             vendor_id: "10de"
    Note

    If the nova-api service is running in a role other than the Controller, then replace ControllerExtraConfig with the user role, in the format <Role>ExtraConfig.

  5. To configure the Compute node on the overcloud for PCI passthrough, create an environment file, for example, pci_passthru_compute.yaml.
  6. To specify the allowed PCIs for the devices on the Compute node, add the following to pci_passthru_compute.yaml:

    parameter_defaults:
      NovaPCIPassthrough:
        - vendor_id: "10de"
          product_id: "1eb8"
  7. To enable IOMMU in the server BIOS of the Compute nodes to support PCI passthrough, add the KernelArgs parameter to pci_passthru_compute.yaml:

       parameter_defaults:
          ...
          ComputeParameters:
            KernelArgs: "intel_iommu=on iommu=pt"
  8. Deploy the overcloud, adding your custom environment files to the stack along with your other environment files:

    (undercloud) $ openstack overcloud deploy --templates \
      -e [your environment files]
      -e /home/stack/templates/pci_passthru_controller.yaml
      -e /home/stack/templates/pci_passthru_compute.yaml
  9. Configure a flavor to request the PCI devices. The following example requests two devices, each with a vendor ID of 10de and a product ID of 13f2:

    # openstack flavor set m1.large --property "pci_passthrough:alias"="t4:2"
  10. Create an instance with a PCI passthrough device:

    # openstack server create --flavor m1.large --image rhelgpu --wait test-pci
  11. Log in to the instance as a cloud-user. For more information, see Log in to an Instance.
  12. Install the GPU driver on the instance. For example, run the following script to install an NVIDIA driver:

    $ sh NVIDIA-Linux-x86_64-430.24-grid.run

Verification

  1. To verify that the GPU is accessible from the instance, run the following command from the instance:

    $ lspci -nn | grep <gpu_name>
  2. To check the NVIDIA System Management Interface status, run the following command from the instance:

    $ nvidia-smi

    Example output:

    -----------------------------------------------------------------------------
    | NVIDIA-SMI 440.33.01    Driver Version: 440.33.01    CUDA Version: 10.2     |
    |---------------------------------------------------------------------------+
    | GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
    | Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
    |===========================================================================|
    |   0  Tesla T4            Off  | 00000000:01:00.0 Off |                    0 |
    | N/A   43C    P0    20W /  70W |      0MiB / 15109MiB |      0%      Default |
    ---------------------------------------------------------------------------
    
    -----------------------------------------------------------------------------
    | Processes:                                                       GPU Memory |
    |  GPU       PID   Type   Process name                             Usage      |
    |=============================================================================|
    |  No running processes found                                                 |
    -----------------------------------------------------------------------------