IBM Turbonomic has recently introduced the capability to define GPU models as "Move Enabled." This enables VMs that utilize these specified GPU models in vCenter (VC) to be seamlessly relocated using MOVE actions, without causing any disruptions.
Background
In today's technology landscape, the usage of GPUs (Graphics Processing Units) is increasingly prevalent, especially in virtual machine (VM) environments. It's becoming common to configure VMs with virtual GPUs (vGPUs) to leverage their powerful processing capabilities. However, when it comes to migrating a VM equipped with a vGPU to another host, several important considerations come into play:
-
GPU Card Model: To ensure a seamless migration of a VM with a vGPU, the destination host must be equipped with GPU cards of the model identical to those in the source host.
- GPU Availability and Compatibility: The GPU chips on the destination host must be available – meaning they are not assigned to any other VM – or, if they are already in use, they should be assigned to a VM that utilizes the same vGPU type. Moreover, these chips should have sufficient remaining GPU memory buffer to accommodate the needs of the migrating VM.
Enhancement
Turbo now supports considerations related to virtual GPUs (vGPUs). In recommending VM placement actions, Turbo evaluates the GPUs installed on both the source and destination hosts, as well as the vGPU types assigned to VMs. It ensures that placement actions are only suggested if the destination host has compatible GPU cards and vGPU types, along with an adequate GPU memory buffer to support the vGPU type of the VM.
You can use Move Enabled GPUs setting to leverage this feature.
Guidance
Host Entity Information Widget
Turbonomic now has the capability to detect GPU cards installed on VC (Virtual Center) hosts. This information is readily accessible in the host's Entity Information Widget.
Installed GPU Devices: This indicates the number of GPU chips installed on the host, along with the model of GPU chips. It's important to note that the number here represents the count of GPU chips, which may differ from the number of GPU cards if a single card integrates multiple GPU chips.
Supported Virtual GPU Types: This section lists the virtual GPU types that the GPU chip can support. The compatibility depends on the specific nature and capabilities of the GPU card. Refer vGPU documentation from Nvidia.
VM Entity Information Widget
In addition to host capabilities, Turbonomic also detects the vGPU types assigned to a VM in VC.
This detail is displayed in the VM's Entity Information Widget.
Virtual GPU Profile: This shows the vGPU types configured on a VM. Please refer vGPU documentation from Nvidia.
Virtual Machines Policy
Under Virtual Machine Policies -> Operational Constraints, you can find the setting
Move Enabled GPUs: Specify a regular expression (regex) for the GPU chip models you wish to enable placement actions on.
In the above example, for a VM equipped with a vGPU, Turbonomic will initiate MOVE actions exclusively if the hosts' GPU cards belong to either the Nvidia A16 series or the Nvidia T4 series.
Additionally, Turbonomic will verify the vGPU types currently in use on the GPU chips of the destination host. If these chips are already assigned to VMs with different vGPU types, or if all the GPU memory buffers on the chips are fully allocated to VMs, Turbonomic will not generate a MOVE action. This is because the move would not be successful under these conditions
Conclusion
In summary, IBM Turbonomic's latest updates significantly enhance VM management in GPU-enhanced environments. By enabling precise consideration over VM migrations and providing detailed insights into GPU and vGPU configurations, these advancements ensure accurate migration recommendations. Turbonomic continues to evolve as an essential tool for managing complex virtual environments, demonstrating its commitment to innovation and user-centric solutions.
#IBMTurbonomic
#Turbonomic