面向训练仿真系统的动态线程池技术研究
文献类型:学位论文
作者 | 林宗明 |
学位类别 | 硕士 |
答辩日期 | 2011-05-31 |
授予单位 | 中国科学院研究生院 |
授予地点 | 北京 |
导师 | 郑昌文 |
关键词 | 动态线程池 线程调度 任务调度 系统负载 实现模型 |
学位专业 | 计算机应用技术 |
中文摘要 | 线程池技术是提升多线程应用程序性能的重要技术,已经广泛地应用在各种网络服务器应用程序、中间件等领域。线程池的研究重心已经从静态线程池转移到动态线程池,然而,如何动态提高其性能(请求的响应速度和系统的吞吐量)和扩大适用范围仍然是个需要深入研究的问题。基于经验的动态线程池技术虽然给出了简单的估算最优线程池尺寸的公式,但该策略需要人工干预,而且操作复杂,容易出错。 基于预测的动态线程池技术通过分析任务请求的历史分布预测需要的线程数,但难以保证预测精确度,且计算量较大。基于反馈的动态线程池技术通过建立反馈机制,根据线程池的性能表现和系统资源限制动态调整池尺寸,但提出的性能度量仍然不够合理。这些已有的策略仅侧重于线程池尺寸对性能的影响,没有考虑任务调度策略和线程调度策略等对性能的影响。因此,需要全面研究影响线程池性能的因素,从而进一步提高线程池的性能。 本论文在综合考虑了线程池尺寸、任务调度策略、线程调度策略、不同的实现模型以及系统负载情况对线程池性能的影响的基础上,提出了一种基于反馈和负载的动态线程池体系结构,并给出了相应的性能度量、线程调度策略、任务调度策略以及性能监测和动态调整算法。 在上述基础上,将提出的线程池策略应用于训练仿真系统中,并与水位线程池策略进行性能比较。实验结果表明,提出的动态线程池策略能更好地响应任务请求的数量和性质的变化,大幅地改善任务处理时间,具有一定的性能优势。 |
英文摘要 | Thread pool is an important technique of performance optimization in multithreaded applications, which has been widely used in web server application, middleware and so on. The study focus of thread pool has transferred from static thread pool to dynamic one, however, how to improve the performance (response speed of requests and throughput of the system), and expand the application scope remains to be discussed further.Experience-based dynamic thread pool presents the simple formula for estimating the optimal size of thread pool, but it needs manual intervention, which is complicated and error-prone. Prediction-based dynamic thread pool predicts the needed size via analyzing the historical distribution of the requests, but it is time-consuming and hard to guarantee prediction accuracy. Feedback-based dynamic thread pool constructs a kind of feedback mechanism to adjust the size dynamically according to the performance of thread pool and the limitation of system resource, but the presented performance metrics are still not reasonable. These strategies only focus on the influence of size on performance, not considering the influence of task scheduling strategy and thread scheduling strategy on performance. Hence, it is necessary to fully study factors which influence the performance of thread pool in order to improve the performance of thread pool further.The thesis presents a kind of feedback-based and load-based dynamic thread pool architecture, which considers the overall influence of size, task scheduling strategy, thread scheduling strategy, different realization patterns and system load on performance. The thesis also gives the corresponding performance metrics, task scheduling strategy, thread scheduling strategy and the algorithm of performance monitoring and adjusting. The proposed strategy has been applied in practice simulation system, and compared with watermark thread pool. The experiment results show that the proposed strategy can response the variation of requests' quantity and quality better, and improve the processing time of requests significantly, having certain performance advantage. |
学科主题 | 计算机应用 |
语种 | 中文 |
公开日期 | 2011-06-11 |
源URL | [http://124.16.136.157/handle/311060/10438] ![]() |
专题 | 软件研究所_综合信息系统技术国家级重点实验室 _学位论文 |
推荐引用方式 GB/T 7714 | 林宗明. 面向训练仿真系统的动态线程池技术研究[D]. 北京. 中国科学院研究生院. 2011. |
入库方式: OAI收割
来源:软件研究所
其他版本
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。