面向云计算实时服务的调度模型与算法研究

项目名称： 面向云计算实时服务的调度模型与算法研究

项目编号： No.61202043

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 滕飞

作者单位： 西南交通大学

项目金额： 25万元

中文摘要： 随着云计算与各行业广泛而深入的结合，多样化的服务对于时效性要求越来越高。当前云计算实时研究多关注于响应速度而忽视截止期保障。由于云平台数据分布化、计算并行化、集群动态化等特点，传统实时调度理论无法应用于云计算场景，妨碍了云服务的拓展。本课题围绕云计算实时调度的核心问题，首先分析多种基于并行数据计算框架云平台的结构和存储特点，提炼其计算流程的共性，建立形式化、可分析的云计算实时调度理论模型。其次，研究解决实时调度可靠性中的三个关键问题：①硬实时作业的可调度性分析和在线接入控制测试方法；②软实时作业服务器的接纳能力和响应速度；③实时容错版本控制和冗余缩减等。最后，研究大规模异构系统中作业完成时间的估计方法，设计面向现实云计算系统具有避错容错功能的自适应多模实时调度器。本课题旨在增强现有云计算系统对实时服务的处理能力，创新并行数据计算框架的模型研究，从而为建立云计算实时调度理论体系奠定基础。

中文关键词： 云计算；实时性；调度；可靠性；并行计算

英文摘要： With cloud computing services spreading in industries, the picky cloud clients expect to be served in real-time. The recent researches related to real-time cloud computing focus on the quick response, not on the hard deadline. Considering that cloud infrastructure applies distributed data storage, parallel computing model and dynamic cluster configuration, traditional real-time scheduling algorithms are not able to solve real-time scheduling problems in cloud environment. It will limit the extension of cloud service. This proposal aims at finding reliable real-time scheduling algorithms for cloud computing. Firstly, a specified scheduling model will be established to describe the real-time cloud services. This model is obtained by analysis of popular parallel processing frameworks and abstraction of their common execution flow. Secondly, three methods that can improve scheduling reliability will be studied deliberately: (1) Schedulability analysis and on-line admission control test for hard real-time jobs; (2) Server capacity and response speed for soft real-time jobs;(3) Version control and redundancy reducing for real-time fault tolerance. Thirdly, an adaptive real-time scheduler will be developed for hands-on purpose, which should have the ability of fault avoidance and fault tolerance in real cloud infrastru

英文关键词： cloud computing；real-time；scheduling；reliability；parallel computing