Dynamic GPU-enabled virtual machine provisioning across cloud providers
A method of provisioning virtual machines (VMs) includes: providing a VM pool that includes a graphics processing unit (GPU)-optimized VM and a non-GPU-optimized VM operating in different clouds. A control plane can receive an indication that a user has submitted a workload request, determine whether a GPU-optimized VM is available and instruct the non-GPU-optimized VM to send the workload to the GPU-optimized VM in a peer-to-peer manner. The GPU-optimized VM computes the workload and returns a result to the requesting VM. The control plane can instantiate a new GPU-optimized VM (or terminate it when the workload is complete) to dynamically maintain a desired number of available GPU-optimized VMs.
1. A method for dynamic virtual machine (VM) provisioning across cloud service providers, comprising:
transmitting instructions across computer networks to hypervisors executing on computers of the cloud service providers to provide a VM pool that includes a first graphics processing unit (GPU)-optimized VM and a first non-GPU-optimized VM;
receiving, at a control plane for the VM pool, an indication that a user of the first non-GPU-optimized VM has submitted a request through the first non-GPU-optimized VM, wherein the request specifies a workload;
determining, by the control plane, that the first GPU-optimized VM is available for using a GPU to execute the workload, and issuing, by the control plane upon said determining, an instruction to the first non-GPU-optimized VM to send the workload to the first GPU-optimized VM; and
transmitting an instruction across a computer network to a computer of one of the cloud service providers to instantiate a second GPU-optimized VM to increase the amount of GPU-optimized VMs of the VM pool that are available for using GPUs to execute workloads, to a minimum number.
2. The method of claim 1 , further comprising:
receiving, at the control plane, an indication that the workload has been completed by the first GPU-optimized VM using the GPU;
determining that the first GPU-optimized VM is available for using GPUs to execute workloads; and
causing at least one of the first and second GPU-optimized VMs to shut down based on instructions in a profile that indicate a maximum number of GPU-optimized VMs to be available for using GPUs to execute workloads.
3. The method of claim 1 , wherein the request is received through a virtual desktop interface provided by the first non-GPU-optimized VM.
4. The method of claim 1 , wherein the first non-GPU-optimized VM and the first GPU-optimized VM execute on computers of different cloud service providers.
5. The method of claim 1 , wherein the first and second GPU-optimized VMs execute on computers of different cloud service providers.
6. The method of claim 1 , wherein the control plane executes on a third GPU-optimized VM.
7. A non-transitory, computer-readable medium containing instructions that, when executed by a hardware-based processor, cause the processor to perform stages for dynamic virtual machine (VM) provisioning across cloud service providers, the stages comprising:
transmitting instructions across computer networks to hypervisors executing on computers of the cloud service providers to provide a VM pool that includes a first graphics processing unit (GPU)-optimized VM and a first non-GPU-optimized VM;
receiving, at a control plane for the VM pool, an indication that a user of the first non-GPU-optimized VM has submitted a request through the first non-GPU-optimized VM, wherein the request specifies a workload;
determining, by the control plane, that the first GPU-optimized VM is available for using a GPU to execute the workload, and issuing, by the control plane upon said determining, an instruction to the first non-GPU-optimized VM to send the workload to the first GPU-optimized VM; and
transmitting an instruction across a computer network to a computer of one of the cloud service providers to instantiate a second GPU-optimized VM to increase the amount of GPU-optimized VMs of the VM pool that are available for using GPUs to execute workloads, to a minimum number.
8. The non-transitory, computer-readable medium of claim 7 , the stages further comprising:
receiving, at the control plane, an indication that the workload has been completed by the first GPU-optimized VM using the GPU;
determining that the first GPU-optimized VM is available for using GPUs to execute workloads; and
causing at least one of the first and second GPU-optimized VMs to shut down based on instructions in a profile that indicate a maximum number of GPU-optimized VMs to be available for using GPUs to execute workloads.
9. The non-transitory, computer-readable medium of claim 7 , wherein the request is received through a virtual desktop interface provided by the first non-GPU-optimized VM.
10. The non-transitory, computer-readable medium of claim 7 , wherein the first non-GPU-optimized VM and the first GPU-optimized VM execute on computers of different cloud service providers.
11. The non-transitory, computer-readable medium of claim 7 , wherein the first and second GPU-optimized VMs execute on computers of different cloud service providers.
12. The non-transitory, computer-readable medium of claim 7 , wherein the control plane executes on a third GPU-optimized VM.
13. A system for dynamic virtual machine (VM) provisioning across cloud service providers, comprising:
a memory storage comprising instructions; and
a computing device including a hardware-based processor that is configured to execute the instructions from the memory storage to carry out stages comprising:
transmitting instructions across computer networks to hypervisors executing on computers of the cloud service providers to provide a VM pool that includes a first graphics processing unit (GPU)-optimized VM and a first non-GPU-optimized VM;
receiving, at a control plane for the VM pool, an indication that a user of the first non-GPU-optimized VM has submitted a request through the first non-GPU-optimized VM, wherein the request specifies a workload;
determining, by the control plane, that the first GPU-optimized VM is available for using a GPU to execute the workload, and issuing, by the control plane upon said determining, an instruction to the first non-GPU-optimized VM to send the workload to the first GPU-optimized VM; and
transmitting an instruction across a computer network to a computer of one of the cloud service providers to instantiate a second GPU-optimized VM to increase the amount of GPU-optimized VMs of the VM pool that are available for using GPUs to execute workloads, to a minimum number.
14. The system of claim 13 , the stages further comprising:
receiving, at the control plane, an indication that the workload has been completed by the first GPU-optimized VM using the GPU;
determining that the first GPU-optimized VM is available for using GPUs to execute workloads; and
causing at least one of the first and second GPU-optimized VMs to shut down based on instructions in a profile that indicate a maximum number of GPU-optimized VMs to be available for using GPUs to execute workloads.
15. The system of claim 13 , wherein the request is received through a virtual desktop interface provided by the first non-GPU-optimized VM.
16. The system of claim 13 , wherein the first non-GPU-optimized VM and the first GPU-optimized VM execute on computers of different cloud service providers.
17. The system of claim 13 , wherein the first and second GPU-optimized VMs execute on computers of different cloud service providers.
18. The system of claim 13 , wherein the control plane executes on a third GPU-optimized VM.