混合现实中基于GPU虚拟化的AI计算优化
Optimization of AI computation in mixed reality based on GPU virtualization
梁桂才 1李玉荣2
作者信息
- 1. 广西机电职业技术学院信息管理中心,广西南宁 530007
- 2. 山西师范大学临汾学院数计系,山西临汾 041000
- 折叠
摘要
研究探讨混合现实(MR)应用中,通过GPU虚拟化优化AI计算,聚焦于多任务调度与资源共享.研究提出了一个模型,其包含一种根据任务优先级、资源需求和等待时间,动态为正在执行的任务分配GPU资源的机制.同时,模型采用优化的多任务调度算法,以提高调度效率.实验结果表明,尽管在单任务性能测试中模型的执行时间、GPU利用率和内存使用方面略逊于物理GPU,但在多任务并发和资源共享方面,研究提出的模型展现了显著优势.未来研究将探索设计更公平高效的资源共享策略,以及进一步优化多任务调度算法.
Abstract
This research explores the optimization of AI computation in Mixed Reality(MR)applications through GPU virtualiza-tion,focusing on multi-task scheduling and resource sharing.A model is proposed in this study that includes a mechanism for dynami-cally allocating GPU resources to tasks in progress based on task priority,resource demand,and wait time.Simultaneously,the model adopts an optimized multi-task scheduling algorithm to enhance scheduling efficiency.Experimental results show that although the model's execution time,GPU utilization rate,and memory usage are slightly inferior to the physical GPU in single-task performance tests,it demonstrates significant advantages in multi-task concurrency and resource sharing.Future research will explore the design of more fair and efficient resource sharing strategies and further optimize the multi-task scheduling algorithm.
关键词
混合现实/AI计算/多任务调度/资源共享/GPU虚拟化Key words
Mixed reality/AI computation/Multi-task scheduling/Resource sharing/GPU virtualization引用本文复制引用
基金项目
2023年广西科技厅广西重点研发计划项目(2023AB01399)
出版年
2024
国家科技期刊平台
NSTL
万方数据