公开(公告)号：US11113782B2

公开(公告)日：2021-09-07

申请号：US16601831

申请日：2019-10-15

Applicant: VMware, Inc.

Inventor： Chandra Prakash ,  Anshuj Garg ,  Uday Pundalik Kurkure ,  Hari Sivaraman ,  Lan Vu ,  Sairam Veeraswamy

IPC: G06T1/20 ,  G06F9/455 ,  G06F9/48

Abstract: Various examples are disclosed for dynamic kernel slicing for virtual graphics processing unit (vGPU) sharing in serverless computing systems. A computing device is configured to provide a serverless computing service, receive a request for execution of program code in the serverless computing service in which a plurality of virtual graphics processing units (vGPUs) are used in the execution of the program code, determine a slice size to partition a compute kernel of the program code into a plurality of sub-kernels for concurrent execution by the vGPUs, the slice size being determined for individual ones of the sub-kernels based on an optimization function that considers a load on a GPU, determine an execution schedule for executing the individual ones of the sub-kernels on the vGPUs in accordance with a scheduling policy, and execute the sub-kernels on the vGPUs as partitioned in accordance with the execution schedule.

公开(公告)号：US20210110506A1

公开(公告)日：2021-04-15

申请号：US16601831

申请日：2019-10-15

Applicant: VMware, Inc.

Inventor： Chandra Prakash ,  Anshuj Garg ,  Uday Pundalik Kurkure ,  Hari Sivaraman ,  Lan VU ,  Sairam Veeraswamy

IPC: G06T1/20 ,  G06F9/48 ,  G06F9/455

Abstract: Various examples are disclosed for dynamic kernel slicing for virtual graphics processing unit (vGPU) sharing in serverless computing systems. A computing device is configured to provide a serverless computing service, receive a request for execution of program code in the serverless computing service in which a plurality of virtual graphics processing units (vGPUs) are used in the execution of the program code, determine a slice size to partition a compute kernel of the program code into a plurality of sub-kernels for concurrent execution by the vGPUs, the slice size being determined for individual ones of the sub-kernels based on an optimization function that considers a load on a GPU, determine an execution schedule for executing the individual ones of the sub-kernels on the vGPUs in accordance with a scheduling policy, and execute the sub-kernels on the vGPUs as partitioned in accordance with the execution schedule.