基于虚拟计算群的众核处理器动态在线任务调度算法研究

发布时间：2018-06-02 03:14

本文选题：众核处理器 + 虚拟计算群　；参考：《上海交通大学》2013年硕士论文

【摘要】：随着微处理器体系结构技术的发展，众核处理器已经成为未来微处理器的一个重要发展方向。如何通过实现高效的任务调度机制提高资源利用率和吞吐率，是当前学术界的研究热点。本课题基于片内集成数十到数百个简单处理器核的众核处理器，面向不同任务的动态性、同一任务的阶段性资源需求，提出基于虚拟计算群的任务调度方案以及基于树的矩形资源管理算法，有效减少了众核处理器核间通信代价，提高简单众核处理器的资源利用率和吞吐率。本文在三个方面做了研究，首先综合考虑核间通信、任务阶段性资源需求等因素提出了基于虚拟计算群的任务调度机制，主要分为动态映射和在线映射两个方面。第二，针对众核处理器系统的矩形结构，提出一种基于树的资源管理算法，有效的优化了核资源分配问题，极大的减少了空闲资源查询和分配带来的时间延迟，并且为将来任务阶段性的占有、释放核资源提供高效的查询分配措施。第三，，在GEM5平台上配置建立众核处理器仿真平台，随机生成大量的任务，对任务进行阶段性划分，应用算法实现任务调度和资源分配，对此进行了有效验证。实验结果与传统多核调度算法相比，本文提出的算法能够有效的减少系统执行时间，提高大约18%的系统吞吐率和23%的资源利用率。
[Abstract]:With the development of microprocessor architecture technology, multi-core processor has become an important development direction of microprocessor in the future. How to achieve efficient task scheduling mechanism to improve resource utilization and throughput is a hot topic in academia. This topic is based on the multi-core processor which integrates dozens to hundreds of simple processor cores, which is oriented to the dynamic nature of different tasks and the phase resource requirements of the same task. A task scheduling scheme based on virtual computing group and a tree-based rectangular resource management algorithm are proposed, which can effectively reduce the communication cost between cores and improve the resource utilization and throughput of simple multi-core processors. In this paper, three aspects are studied. Firstly, the task scheduling mechanism based on virtual computing cluster is proposed, which includes dynamic mapping and online mapping. Secondly, for the rectangular structure of multi-core processor system, a tree-based resource management algorithm is proposed, which effectively optimizes the kernel resource allocation problem and greatly reduces the time delay caused by the query and allocation of idle resources. It also provides efficient query and allocation measures for future task stage possession and release of nuclear resources. Thirdly, the multi-core processor simulation platform is configured on the GEM5 platform, and a large number of tasks are generated randomly, the tasks are partitioned into phases, and the task scheduling and resource allocation are implemented by the algorithm, which is validated effectively. Compared with the traditional multi-core scheduling algorithm, the proposed algorithm can effectively reduce the system execution time, improve the system throughput by about 18% and resource utilization by 23%.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP332

【共引文献】