Description
Improve vGPU profile scheduling and allocation to properly handle hosts with multiple physical GPUs.
Enhance scheduler functionality:
- allow placement rules for vGPU's, packing/striping for systems with multiple GPUs
- increase awareness of GPU capacities and usage with currently deployed profiles.
Use case
When multiple vGPUs are allocated on a host with a multiple physical GPU, they may contend with each other instead of being distributed across available devices.
By declaring in advance how each GPU is partitioned and determining which PCI slot ranges belong to each device, the scheduler can distribute VMs utilizing vGPUs more effectively, avoid collisions, and maximize overall utilization.
Interface Changes
VM / VM Template: add fields for vGPU profile on the PCI device
Host View: Display GPUs and capacity with possible profiles, assign profiles per-GPU.
Progress Status
Description
Improve vGPU profile scheduling and allocation to properly handle hosts with multiple physical GPUs.
Enhance scheduler functionality:
Use case
When multiple vGPUs are allocated on a host with a multiple physical GPU, they may contend with each other instead of being distributed across available devices.
By declaring in advance how each GPU is partitioned and determining which PCI slot ranges belong to each device, the scheduler can distribute VMs utilizing vGPUs more effectively, avoid collisions, and maximize overall utilization.
Interface Changes
VM / VM Template: add fields for vGPU profile on the PCI device
Host View: Display GPUs and capacity with possible profiles, assign profiles per-GPU.
Progress Status