HeteroPrio: An Efficient Heterogeneous Computing Resource Scheduling Framework

This paper presents HeteroPrio, a heterogeneous computing resource scheduling framework designed for efficient task scheduling and resource management in heterogeneous computing environments. HeteroPrio exploits the heterogeneity of diverse computing resources (such as CPUs, GPUs, and FPGAs) by dynamically allocating tasks to the most appropriate resources based on task properties and resource availability. Furthermore, HeteroPrio dynamically adjusts resource allocation priorities based on task urgency or importance, ensuring timely task completion and optimal resource utilization.

The HeteroPrio framework consists of three core components: a scheduler, a resource manager, and a priority manager. The scheduler is responsible for receiving and assigning tasks, dynamically allocating them to the most suitable computing resources based on task requirements and resource availability. The resource manager monitors and manages the state and utilization of computing resources, coordinating resource allocation and reclamation. The priority manager dynamically adjusts resource allocation priorities based on task urgency and importance, ensuring timely task completion and optimal resource utilization.

Experimental results demonstrate that HeteroPrio effectively enhances task execution efficiency and resource utilization in heterogeneous computing environments while exhibiting high scalability and flexibility. This framework holds significant practical value, finding applications in diverse heterogeneous computing scenarios such as deep learning and big data processing.