Hardware Interrupt and CPU Contention aware CPU/GPU Co-Scheduling on Multi-Cluster System

Abstract

As a co-processor for data-parallel processing and graphics, the GPU plays an important role in recent platforms. GPU is regarded as an input/output device on the CPU-GPU heterogeneous board and generates a hardware interrupt to communicate with the CPU. Since the CPU delays the execution of the GPU workload to process such a hardware interrupt, the hardware interrupt is a factor that degrades the GPU performance. In this paper, we propose a method to improve the processing performance of GPU by separating the CPU core that handles hardware interrupts from the CPU core that handles GPU tasks in the CPU-GPU heterogeneous architecture. In addition, CPU contention was resolved by separating the CPU core that processes the CPU task from the GPU task.