一种IaaS模式“云训练”系统设计  

Design of IaaS Mode “Cloud Training” System

作　　者：陈志佳[1] 朱元昌[1] 邸彦强[1] 冯少冲[1] 

机构地区：[1]军械工程学院电子与光学工程系,石家庄050003

出　　处：《系统仿真学报》2015年第5期1095-1104,共10页Journal of System Simulation

基　　金：装备预研基金(9140A04030214JB34001)

摘　　要：针对网络化模拟训练真实感不强,训练效率低等问题,提出了一种Iaa S(Infrastructure as a Service)模式的"云训练"。阐述了"云训练"的体系结构和运行模式,研究了"云训练"的3项核心技术。通过GPU虚拟化技术解决云环境中终端用户3D图形图像处理能力弱的问题。根据模拟训练特点,建立用户需求模型,将模糊理论引入资源调度技术中,实现虚拟资源的动态调度;结合检查点回滚、虚拟机备份和虚拟机迁移技术,实现系统高效动态容错,保证了系统的稳定性与可靠性。实验证明,"云训练"系统改善了传统模拟训练系统中资源的按需共享能力,提升了资源利用率,保证了模拟训练效果。There are some problems in equipment web simulating training, including poor sense of reality, low efficiency of training and high difficulty of management. To solve those problems, Infrastructure as a Service mode 'cloud training' was proposed. The architecture and operation flow were explained in detail and the key technologies of 'cloud training' were focused on. By GPU virtualization technology, the problem of client 3D image processing capability was solved. According to the characters of simulation training, the user requirement model was established. Fuzzy algorithm was introduced to realize resource dynamic scheduling. To assure the stability and reliability, combined with checkpoint, virtual machine backup and migration, dynamic failure tolerance algorithm was given. The results show that resource-sharing capability and simulating training effect have been improved.

关 键 词：模拟训练 IaaS(基础设施即服务) 云训练 GPU虚拟化 资源动态调度 动态容错 

分 类 号：TP393.09[自动化与计算机技术—计算机应用技术] TP391.9[自动化与计算机技术—计算机科学与技术]