Skip to main content

Manage intelligent resource scheduling

Intelligent resource scheduling (iRS) adds resources on different hosts in the same cluster to a resource

group and adds VMs that provide the same service to a VM group. A service template defines the priority of VMs that use the service template to use physical resources and the total ratio of resources that all VMs using the low-priority service template can use. When a VM starts or restarts, the management platform allocates resources to the VM based on its service template priority, resource usage of the resource group, and the total ratio of resources that all VMs using the same service template use. Only iRS services for GPU and vGPU and SR-IOV resources are supported in the current software version. Before configuring iRS services, configure service templates as needed.

Features

• Manage service templates

• Manage iRS services

• View VM information in an iRS service

• View performance monitoring information in an iRS service

• View resource details in an iRS service

• View resource usage history in an iRS service

Manage service templates​

A service template defines the rules for VMs to use the physical resources on a host, such as GPUs. The

management platform assigns resources based on the priorities defined in the service templates when resources are insufficient. A VM with higher priority can use the corresponding resources first. By default, a predefined non-linear editing service template and a transcoding service template exist in the system. You can add more service templates as needed.

Restrictions and guidelines

Service templates used by VMs cannot be deleted.

Add a service template

  1. On the top navigation bar, click Compute.

  2. From the left navigation pane, select iRS.

  3. Click Service Template.

  4. Click Add.

  5. Configure the parameters as described in "Parameters."

  6. Click OK.

Edit a service template

  1. On the top navigation bar, click Compute.

  2. From the left navigation pane, select iRS.

  3. Click Service Template.

  4. Click Edit in the Actions column for a service template.

  5. Configure the parameters as described in "Parameters."

  6. Click OK.

Delete a service template

  1. On the top navigation bar, click Compute.

  2. From the left navigation pane, select iRS.

  3. Click Service Template.

  4. Click Delete in the Actions column for a service template.

  5. In the dialog box that opens, click OK.

View service template details

  1. On the top navigation bar, click Compute.

  2. From the left navigation pane, select iRS.

  3. Click Service Template.

  4. Click View for a template to view its details.

Parameters

• Priority: The management platform adjusts resource assignment based on this priority if the host does not have resources that can be assigned to VMs. High-priority VMs can be assigned resources first. Low-priority VMs might be preempted.

• VM Startup Mode: When the resources in a GPU or vGPU resource pool are insufficient, you can set a startup mode for VMs that have not been assigned any resources. Options include the following:

○ Not Start—The system does not start a VM if no resources are available or the available resources in the resource pool cannot meet the requirement of the VM.

○ Start Without Resources—The system starts a VM without assigning any resources to the VM if no resources are available in the resource pool. If the available resources in the resource pool cannot meet the requirement of the VM, the system assigns the available resources to the VM and start the VM. For example, if two GPUs are available in the resource pool, but the VM requires three GPUs, the system will start the VM and assign the two GPUs to the VM.

○ Preempt—High-priority VMs can preempt low-priority VMs and use the resources that have been assigned to the low-priority VMs.

• Preemption Threshold: When this threshold is exceeded, high-priority VMs will try to preempt low-priority VMs if the VM startup mode is set to Preempt, in percent. Assume that a GPU resource pool has 10 GPUs and the threshold is set to 20% (2 GPUs), and iRS assigns 5 GPUs (50%) to the VMs that use the service template. When high-priority VMs start up, but the GPU resource pool does not have available resources, high-priority VMs will try to preempt low-priority VMs. The sum of the preemption thresholds configured in all the service templates in the system cannot exceed 100%. This parameter is required when the VM startup mode is set to Preempt.

• Service Stop Command: Command that can be executed by the OS of a VM to release the resources configured with iRS on the VM so that other VMs can use the resources, for example, shutdown. The system will try to release the GPU/vGPU resources used by a VM only if this parameter is configured correctly. This parameter is required when the VM startup mode is set to Preempt.

• Return Result: Used by the management platform to determine whether a stopping service command has been executed successfully by matching the returned result against this parameter. If the result is success, the VMs release resources. If the result is failure, the system takes the

Manage iRS services​

Configure iRS services to enable the management platform to dynamically allocate resources to VMs based

on the service template priority, resource usage of the resource group, and the allocated resource ratio. For example, VMs directly use GPUs on hosts, and a GPU can be used by only one VM at a time. If the host of a VM does not have available GPU resources, the system migrates the VM to another host that have free GPU resources. When the GPU resources of a cluster are insufficient, VMs using the service template with the high priority take precedence over VMs using the service template with the low priority on GPU resources. If the total ratio of GPU resources used by VMs using the service template with the low priority exceeds the specified allocation ratio, the system reclaims the GPU resources of VMs with the low priority to ensure that VMs with the high priority have sufficient GPU resources to use.

Add or edit an iRS service

  1. On the top navigation bar, click Compute.

  2. From the left navigation pane, select iRS.

  3. To add an iRS service, click Add iRS Service.

  4. To edit an iRS service, click Edit in the Actions column for that iRS service.

  5. Configure basic settings, and then click Next.

    ○ Service Name: Specify the name of the iRS service.

    ○ Description: Specify a description for the iRS service.

    ○ Cluster: Enter the name of the cluster that runs the iRS service.

    ○ Resource Type: Select a resource type. When ARM hosts exist in the cluster or the management platform is installed with the Tencent OS, only the GPU and SR-IOV resource types are supported.

    • You can select vGPU resource types only after you install an enterprise edition license on the management platform. For information about license registration, see NSphere Local Licensing Guide and Software Products Remote Licensing Guide.

    • When the resource type is vGPU, vGPU resources in the resource group must be of the same type.

    • Virtualized GPU resources cannot be added to the GPU resource group.

    • To allocate multiple GPUs to a VM, you must specify GPUs from the same resource pool.

  6. Select the resource to be added, and then click Next.

  7. Select the VMs to be associated, and then click OK. This step is not available for SR-IOV resources. Images files of VMs of an iRS service must be saved on a shared storage resource, and the hosts to which the iRS service resources belong must have mounted the shared storage resource. If the image file of a VM is not saved on the shared storage resource, the VM can use only physical resources of the host to which the VM belongs.

    ○ Service Template: Service template to be used by the VM.

    ○ Host: Host where the VM resides.

    ○ Driver: Driver type of the resource that the VM is using.

    ○ Exclusive Mode: Select whether the VM can exclusively use the specified GPU/vGPU resources. If you select Yes, the GPU/vGPU resources cannot be used by any other VMs. This feature is available only when the host has available GPU/vGPU resources that have been added to the selected GPU resource pool. To migrate a VM when this feature is enabled, make sure sufficient GPU/vGPU resources are available on the target host. A VM supports online migration if it uses vGPU resources. A VM that uses GPU resources can be migrated only when it is shutdown.

    ○ Resource Count: Set the maximum number of GPU/vGPU resources that can be used by the VM. The value depends on the resource pool type and state of the exclusive mode feature:

    ▪ If you select a vGPU resource pool, only 1 is available.

    ▪ If you select a GPU resource pool and exclusive mode is disabled, the value is the maximum number of GPUs on a single host in the resource pool. For example, if the resource pool has three hosts, and the hosts have three, two, and two GPUs, respectively, the value is 3.

    ▪ If you select a GPU resource pool and exclusive mode is enabled, the value is the number of available GPUs on the host attached to the VM.

Delete an iRS service

  1. On the top navigation bar, click Compute.

  2. From the left navigation pane, select iRS.

  3. Click Delete in the Actions column for an iRS service.

  4. In the dialog box that opens, click OK.

View VM information in an iRS service​

Perform this task to display VM information in an iRS service.

Procedure

  1. On the top navigation bar, click Compute.

  2. From the left navigation pane, select iRS.

  3. Click the name of an iRS service to view VM information.

    ○ Alias: Alias of the VM.

    ○ State: State of the VM.

    ○ CPU Usage: Average CPU usage of the VM in the last one hour.

    ○ Memory Usage: Average memory usage of the VM in the last one hour.

    ○ OS: Operating system that runs on the VM. Only Windows and Linux are supported. By default, this column is hidden.

    ○ Service Template: Service template used by the iRS service. You can add service templates as needed.

    ○ Driver Type: Driver type of the VM.

    ○ Slot Number: Slot number of the resource that the VM is using.

    ○ Host: Host where the VM resides.

    ○ Exclusive Mode: Whether the VM can exclusively use the GPU/vGPU resources. With this feature enabled, the GPU/vGPU resources are bound to the VM even if the VM is shut down.

View performance monitoring information in an iRS service​

With this feature disabled, the GPU/vGPU resources can be released after the VM is shut

down.

  1. To edit an iRS service, click Edit iRS Service.

  2. To add a resource to the iRS associated with the VM, click Add Resource.

View performance monitoring information in an iRS service

Perform this task to display performance monitoring information for an iRS service.

Procedure

  1. On the top navigation bar, click Compute.

  2. From the left navigation pane, select iRS.

  3. Click the name of an iRS service.

  4. Click the Performance Monitoring tab.

Parameters

• Usage: GPU/vGPU usage trend within the most recent 30 minutes for the VMs in the iRS service. The horizontal axis represents time and the vertical axis represents GPU/vGPU usage.

View resource details in an iRS service​

Perform this task to view resource details in an iRS service.

  1. On the top navigation bar, click Compute.

  2. From the left navigation pane, select iRS.

  3. Click the name of an iRS service.

  4. Click the Resource Details tab.

    ○ UUID: UUID of the vGPU. This parameter is available only for vGPUs.

    ○ Name: Name of the vGPU. This parameter is available only for vGPUs.

    ○ Monitor Interfaces: Maximum number of monitor interfaces supported by the vGPU. This parameter is available only for vGPUs.

    ○ Buffer: Frame buffer size of the vGPU. This parameter is available only for vGPUs.

    ○ Resolution: Maximum resolution supported by the vGPU. This parameter is available only for vGPUs.

    ○ Slot Number: Slot number of the resource.

    ○ Manufacturer: Manufacturer of the resource. This parameter is available only for GPUs.

    ○ SR-IOV NIC Address: SR-IOV NIC address used by the resource. This parameter is available only when the resource type is SR-IOV.

    ○ Physical NIC: Physical NIC used by the resource. This parameter is available only when the resource type is SR-IOV.

    ○ Physical Interface Model: Physical NIC model used by the resource. This parameter is available only when the resource type is SR-IOV.

    ○ Host: Host where the resource is located.

    ○ State: State of the resource.

    ○ VM: VM that is using the resource. This column is empty if the resource state is Free.

  5. To remove a single resource, click Remove in the Actions column for that resource, and then click OK.

  6. To remove multiple resources in bulk, select the target resources, click Bulk Remove, and then click OK.

View resource usage history in an iRS service​

Perform this task to display resource usage history in an iRS service.

  1. On the top navigation bar, click Compute.

  2. From the left navigation pane, select iRS.

  3. Click the name of an iRS service.

  4. Click the Resource Usage History tab.

    ○ UUID: UUID of the vGPU. This parameter is available only for vGPUs.

    ○ Name: Name of the vGPU. This parameter is available only for vGPUs.

    ○ Monitor Interfaces: Maximum number of monitor interfaces supported by the vGPU. This parameter is available only for vGPUs.

    ○ Buffer: Frame buffer size of the vGPU. This parameter is available only for vGPUs.

    ○ Resolution: Maximum resolution supported by the vGPU. This parameter is available only for vGPUs.

    ○ Slot Number: Slot number of the resource.

    ○ Manufacturer: Manufacturer of the resource. This parameter is available only for GPUs.

    ○ SR-IOV NIC Address: SR-IOV NIC address used by the resource. This parameter is available only when the resource type is SR-IOV.

    ○ Physical NIC: Physical NIC used by the resource. This parameter is available only when the resource type is SR-IOV.

    ○ Physical Interface Model: Physical NIC model used by the resource. This parameter is available only when the resource type is SR-IOV.

    ○ Host: Host where the resource is located.