并发多播队列的实现框架及其多种实现的性能分析

发布时间：2018-05-09 21:09

本文选题：多播队列 + 并发数据结构　；参考：《小型微型计算机系统》2017年06期

【摘要】：开发易用且高效的并发数据结构对降低并行编程的难度和有效利用并行资源非常重要.针对所提出的易于编程的确定性消息传递多线程编程模型DetMP,除可以基于所提出的单生产多播共享虚拟内存模型(SPMC)实现以外,还可以基于传统的多线程共享虚拟内存模型来实现.为了分析消息通道的实现机制(如数据的存储组织、并发访问的同步控制)对DetMP程序性能的影响,提出一个并发多播队列的框架CMQue,并基于Pthreads实现了6种并发多播队列.我们评估了6种并发多播队列和SPMC通道,结果表明消息通道的实现机制对程序性能影响很大,SPMC通道在CPU核资源充足时具有很好的可伸缩性.
[Abstract]:It is very important to develop easy and efficient concurrent data structures to reduce the difficulty of parallel programming and to utilize parallel resources effectively. For the deterministic message passing multithreaded programming model DetMPwhich is easy to program, it can be implemented not only based on the single production multicast shared virtual memory model (SPMCMC), but also based on the traditional multithreaded shared virtual memory model. In order to analyze the influence of the implementation mechanism of message channel (such as data storage organization, synchronization control of concurrent access) on the performance of DetMP program, a framework of concurrent multicast queue (CMQuee) is proposed, and six kinds of concurrent multicast queues are implemented based on Pthreads. We evaluate six kinds of concurrent multicast queues and SPMC channels. The results show that the implementation mechanism of message channels has a great impact on program performance. SPMC channels have good scalability when the CPU core resources are sufficient.
【作者单位】：中国科学技术大学计算机科学与技术学院;
【基金】：国家自然科学基金项目(61229201)资助国家“八六三”高技术研究发展计划项目(2012AA010901)资助
【分类号】：TP311.1;TP333

【相似文献】