一种基于PVFS的并行核外计算方法

A parallel out-of-core computing method using PVFS

针对传统的并行核外编程复杂、虚存系统性能不理想的问题,提出一种基于PVFS的并行核外计算方法,描述了其软件层次结构、数据访问策略及实现,为工作站机群用户提供了方便的编程接口以及核外数据的全局视图.实验结果表明,该方法扩展了可有效解决的核外问题规模,大数据量时的计算性能优于虚存系统.

Since the traditional parallel out-of-core programs are difficult to write and the virtual memory system does not perform well, a parallel out-of-core computing method using PVFS is presented, which provides convenient interface to write out-of-core codes and the global view of the out-of-core data on cluster of worksta- tions. The software level architecture, the data access strategy and implementation are described. Experimental results show that this method extends the problem sizes that can be efficiently solved and its performance is better than the virtual memory system while the data set is large.