摘要
大数据量应用问题引入核外计算模式,由于访问磁盘数据的速度比较慢,I/O成为核外计算性能重要的限制因素·提出了一种使用运行库进行I/O优化的方法,给出了3种有效的优化策略:规则区域筛选、数据预取和边缘重用·编程人员可针对不同的应用问题使用相应的优化API来缩短程序执行时间·实验结果表明,通过减少I/O操作次数和内外存交换的数据量以及隐藏部分I/O操作延迟,有效提高了核外计算的性能·
Applications with large amounts of data bring the mode of out-of-core computation in which I/O becomes the important limiting factor because of the low speed of accessing data on disks. A method of using runtime library is presented for I/O optimizations. Three optimization strategies including data sieving on regular section, data prefetching and data reuse on the edge are described. Programmers may adopt corresponding APIs for different applications to reduce the execution time. The experiment results show that the performance of the out-of-core computation is efficiently improved by reducing the number of I/O operations and the amount of exchanged data between the main memory and disks as well as hiding part of the I/O operation latency.
出处
《计算机研究与发展》
EI
CSCD
北大核心
2005年第10期1820-1825,共6页
Journal of Computer Research and Development
基金
国家自然科学基金项目(90412001)
关键词
核外计算
规则区域筛选
预取
边缘重用
out-of-core computation
regular section sieving
prefetching
data reuse on the edge
