面向Julia的云计算技术研究

发布时间：2018-11-07 19:18

【摘要】：高性能计算机以其强大的计算能力，成为科学与工程计算的有力工具。在中小企业的创新计算过程中，需要共享使用超级计算机，却因编程复杂而难以实施。Julia语言是一门新出现的开源脚本语言，从语言级支持高性能计算，易于编写高性能程序。云计算技术为优秀的计算资源共享模式。本文尝试构建了面向Julia的云计算平台，降低企业用户编写调试并行程序难度的同时，实现多用户共享超级计算资源。本文的主要工作有： (1)面向Julia语言的云服务平台框架：分析Julia语言的并行特性，结合云服务特点，设计了包含前端系统(程序编辑环境、用户管理、作业管理)、后端系统(会话控制、作业调度、状态管理)和消息通信系统三个子系统的高性能云服务平台框架。 (2)基于消息中间件的消息通信系统：定义了可扩展文本消息格式，设计了基于ActiveMQ服务的消息通信机制。通过实现该消息机制，降低了前端系统和后端系统的耦合性，使系统能适应异构环境，，便于构建高可扩展的高性能云服务平台。 (3)云服务平台的后端系统：采用虚拟化技术搭建了一个并行计算平台，并完成相应环境配置。在此基础上，针对并行编程和资源管理，初步设计、实现了高层次抽象编程模型和作业管理系统。 (4)公交线路走行时长统计的计算案例：基于云服务平台，实现了公交线路走行时长统计的计算案例。计算案例中对比了串行算法和并行优化算法在本项目的云服务平台上的运行效率。实验证明，Julia程序从串行模式到并行模式仅添加较少指令，程序短小精炼。优化算法的执行时间缩短了70%。面向Julia的高性能云服务平台为企业创新提供了易用的高性能程序编程环境，构建了企业用户与高性能计算资源间的桥梁。
[Abstract]:High-performance computer has become a powerful tool for scientific and engineering calculation with its powerful computing power. In the process of innovative computing in small and medium-sized enterprises, it is necessary to share and use supercomputers, but it is difficult to implement because of the complexity of programming. Julia language is a new open source scripting language, which supports high performance computing from the language level and is easy to write high performance programs. Cloud computing technology is an excellent computing resource sharing model. This paper attempts to construct a cloud computing platform for Julia to reduce the difficulty for enterprise users to write and debug parallel programs and to realize multi-user sharing of supercomputing resources at the same time. The main work of this paper is as follows: (1) the framework of cloud service platform oriented to Julia language: analyzing the parallel characteristics of Julia language, combining the characteristics of cloud service, designing the front-end system (program editing environment, user management, job management). The high performance cloud service platform framework of back-end system (session control, job scheduling, state management) and message communication system. (2) message communication system based on message middleware: the extensible text message format is defined and the message communication mechanism based on ActiveMQ service is designed. By implementing the message mechanism, the coupling between front-end system and back-end system is reduced, the system can adapt to heterogeneous environment, and it is convenient to construct a high scalable cloud service platform with high performance. (3) the back-end system of cloud service platform: a parallel computing platform is built with virtualization technology, and the corresponding environment is configured. On this basis, a high-level abstract programming model and a job management system are designed for parallel programming and resource management. (4) calculation case of bus route travel time statistics: based on cloud service platform, the calculation case of bus route travel time statistics is realized. The operation efficiency of serial algorithm and parallel optimization algorithm on cloud service platform is compared. Experiments show that the Julia program only adds fewer instructions from serial mode to parallel mode, and the program is short and refined. The execution time of the optimization algorithm is shortened by 70. The high performance cloud service platform for Julia provides an easy to use high performance programming environment for enterprise innovation and builds a bridge between enterprise users and high performance computing resources.
【学位授予单位】：石家庄铁道大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP312.1;TP393.09

【参考文献】