RESOURCE MANAGEMENT METHOD USING INTERLOCKING BETWEEN APPLICATION STATE AND ORCHESTRATION SERVICE IN GRAPHIC PROCESSING UNIT CLOUD ENVIRONMENT AND SYSTEM THEREOF

摘要

The present invention relates to a service and system scheduled to allocate graphic processing unit (GPU) resources to an application, which uses a cloud orchestration service in a GPU-based cloud environment to be executed in a virtual machine, only in a required state so as to efficiently and flexibly use a high-cost GPU resources. According to the present invention, the service calls a peripheral component interconnect (PCI) hot-plug addition function to allocate a GPU to the virtual machine without downtime when the application requires the GPU by using a state of an application which is provided in a PCI pass-through manner and uses the cloud orchestration service in the GPU-based cloud environment to be executed in a virtual machine, and the service calls a PCI hot-plug cancel function to release the GPU from the virtual machine without downtime when the application does not require the GPU.;COPYRIGHT KIPO 2019