多核机群上数据密集型应用并行程序性能优化

摘要

在异构多核机群系统上利用数据任务块的动态调度策略和全锁定技术,给出一种面向数据密集型应用的结点内主存和可用的共享二级缓存大小中动态调度数据块的多进程级和多线程级并行编程机制,给出了优化数据密集型应用并行程序性能的策略和技术.在多核计算机组成的异构机群上并行求解随机序列多关键字查找的实验结果表明,所给出的多核并行程序设计机制和性能优化方法可行和高效.%Using dynamic data task block scheduling policy and all-locking technology on the heterogeneous multi-core clusters, this paper presents a hybrid parallel programming mechanism of multiprocess-level and multithreaded-level for the data-intensive applications, which can efficiently use the data in the main memory and dynamic schedule the data block in shared L2 cache, and presents the technology and strategy of paralleled application performance optimization for the data-intensive applications. The experiments for solving the multi-keyword search of random sequences parallelly on the heterogeneous multi-core clusters show that the parallel programming mechanism and performance optimization methods are usable and efficient.