Back to results
US 12,608,227 B2Granted
Job allocations to graphics processing units with tenant isolation
AssigneeHewlett Packard Enterprise Development LP
InventorsDiman Zad Tootaghaj, Yunming Xiao, Aditya Dhakal, Puneet Sharma
Life of this patent
Estimating the term…
Abstract
In some examples, a system including physical graphics processing units (GPUs) receives a request to schedule a new job to be executed in the system that is accessible by a plurality of tenants to use the physical GPUs. The system allocates the new job to a collection of vGPUs of the physical GPUs based on an operational cost reduction objective to reduce a cost associated with a usage of the physical GPUs and based on a tenant isolation constraint to provide tenant isolation wherein a single tenant of the plurality of tenants is to use a physical GPU at a time.
Claims18
A non-transitory machine-readable storage medium comprising instructions, which upon execution, cause a system to:
- receive, in the system that includes physical graphics processing units (GPUs), a request to schedule a new job to be executed in the system that is accessible by a plurality of tenants to use the physical GPUs, the request being from a first tenant of the plurality of tenants, and wherein each of the physical GPUs is divided into a respective number of virtual GPUs (vGPUs);
- receive a first weight for an operational cost reduction objective representing a weighted operational cost and associated with reducing a cost of usage of the physical GPUS;
- receive a second weight for a migration cost reduction objective representing a weighted migration cost and associated with reducing a cost of migrating jobs between the physical GPUs;
- and
- execute the new job of the first tenant on a collection of vGPUs of a first physical GPU according to an allocation of the new job while maintaining tenant isolation between the new job on the first physical GPU and another job of a second tenant on a different physical GPU, wherein the allocation is based on the first weight of the operational cost reduction objective, the second weight of the migration cost reduction objective, and a tenant isolation constraint to provide the tenant isolation wherein a single tenant of the plurality of tenants is to use a physical GPU at a time, and wherein the allocation reduces the weighted operational cost and the weighted migration cost while satisfying the tenant isolation constraint.
2 more elements
A computing system comprising:
- a plurality of physical graphics processing units (GPUs), wherein each physical GPU of the plurality of physical GPUs is divided into a respective number of virtual GPUs (vGPUs);
- a hardware processor;
- and
- a non-transitory storage medium storing instructions executable on the hardware processor to:
- receive a request to schedule a new job to be executed in the computing system that is accessible by a plurality of tenants to use the plurality of physical GPUs, the request being from a first tenant of the plurality of tenants;
- receive a first weight for an operational cost reduction objective representing a weighted operational cost and associated with reducing a cost of usage of the physical GPUs;
- receive a second weight for a migration cost reduction objective representing a weighted migration cost and associated with reducing a cost of migrating jobs between the physical GPUs;
- and
- execute the new job of the first tenant on a collection of vGPUs of a first physical GPU according to an allocation of the new job while maintaining tenant isolation between the new job on the first physical GPU and another job of a second tenant on a different physical GPU, wherein the allocation is based on the first weight of the operational cost reduction objective, the second weight of the migration cost reduction objective, and a tenant isolation constraint to provide the tenant isolation wherein a single tenant of the plurality of tenants is to use a physical GPU at a time, and wherein the allocation reduces the weighted operational cost and the weighted migration cost while satisfying the tenant isolation constraint.
6 more elements
A method comprising:
- receiving, in a computing system that includes physical graphics processing units (GPUs), a request to schedule a new job to be executed in the computing system that is accessible by a plurality of tenants to use the physical GPUs, the request being from a first tenant of the plurality of tenants, and wherein each of the physical GPUs is divided into a respective number of virtual GPUs (vGPUs);
- receiving a first weight for an operational cost reduction objective representing a weighted operational cost and associated with reducing a cost of usage of the physical GPUS;
- receiving a second weight for a migration cost reduction objective representing a weighted migration cost and associated with reducing a cost of migrating jobs between the physical GPUs;
- and
- executing the new job of the first tenant on a collection of vGPUs of a first physical GPU according to an allocation of the new job while maintaining tenant isolation between the new job on the first physical GPU and another job of a second tenant on a different physical GPU, wherein the allocation is based on the first weight of the operational cost reduction objective, the second weight of the migration cost reduction objective, and a tenant isolation constraint to provide the tenant isolation wherein a single tenant of the plurality of tenants is to use a physical GPU at a time, and wherein the allocation reduces the weighted operational cost and the weighted migration cost while satisfying the tenant isolation constraint.
2 more elements
Dependencies are inferred from the claim text.