GPU云资源预约式智能调度技术研究  

Research on Intelligent Scheduling of GPU Cloud Resources with Reservation-based System

作　　者：张云鹏 臧亚楠 田秋雨 ZHANG Yunpeng;ZANG Yanan;TIAN Qiuyu(Joint Operations College of National Defense University,Shijiazhuang 050000;Zhengzhou University,Zhengzhou 450000;Nanjing Institute of InforSuperBahn,Nanjing 210000)

机构地区：[1]联合作战学院,石家庄050000 [2]郑州大学,郑州450000 [3]中科南京信息高铁研究院,南京210000

出　　处：《舰船电子工程》2025年第2期119-123,共5页Ship Electronic Engineering

基　　金：江苏省重大科研设施预研筹建项目(编号:BM2021800)资助。

摘　　要：针对GPU云资源的供需不平衡、实时调度策略存在局限性的问题,提出了一种GPU云资源预约式智能调度技术(Deadline Ant Clony Optimization,DACO),该技术由预测时长模块和调度模块两部分组成。文章对提出的技术进行了实验模拟验证,结果表明论文提出的技术能够有效地在保证任务截止时间前完成率的条件下,提高资源利用率和任务响应时间。A GPU cloud resource reservation based intelligent scheduling technology(DACO)is proposed to address the imbalance between supply and demand of GPU cloud resources and the limitations of real-time scheduling strategies.This technology consists of two parts,which are a prediction duration module and a scheduling module.The proposed technology is tested by Analog verification,and the results show that the proposed technology can effectively improve resource utilization and task response time under the condition of ensuring the completion rate before the deadline.

关 键 词：GPU调度 蚁群算法 机器学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]