面向非易失性内存的高性能计算并行优化研究
Research on Parallel Optimization of High Performance Computing for Non volatile Memory
辛华1
作者信息
- 1. 南方电网数字电网研究院股份有限公司,广东广州 510663
- 折叠
摘要
针对非易失性内存在高性能计算运行过程中存在的内存分配不均匀问题,研究非易失性内存的高性能计算并行优化方法.结合新型高速网络 RDMA和非易失性存储设备,提出分布式非易失性内存新型存储结构,加入一致性模块和本地非易失性内存管理模块,同时加强本地非易失性内存的管理;通过SPL框架进行分布式并行计算,经过数据预加载后,使用分布式独立内存分配算法优化多任务多线程的并行计算过程.实验结果表明:该方法能够使非易失性内存的数据保持一致,增强读写性能,提升工作效率;对于并行多线程任务的内存分配具有更好的扩展性,增强系统并行运算能力,同时适用性广泛.
Abstract
To study the problem of non-volatile memory computing parallel optimization method for non-volatile memory.Combined with the new high-speed network RDMA and non-volatile storage equipment,a new storage structure of distributed non-volatile memory is proposed,including the consistency module and local non-volatile memory management module,and strengthening the management of local non-volatile memory;conduct distributed parallel computing through SPL framework,and optimize the parallel computing process of multi-task and multi-threading after data preloading by distributed independent memory allocation algorithm.The experimental results show that this method can keep the data of non-volatile memory consistent,enhance the read and write performance,improve the work efficiency;expand the memory allocation of parallel multi-threaded tasks,enhance the parallel computing ability of the system,and have wide applicability.
关键词
非易失性内存/高性能计算/并行优化/分布式储存结构/内存分配算法/CPU资源分配Key words
Non volatile memory/High performance computing/Parallel optimization/Distributed storage structure/Memory allocation algorithm/CPU resource allocation引用本文复制引用
基金项目
南方电网科技项目(JY-YZ-02-CJ-22-003)
出版年
2024
国家科技期刊平台
NSTL
万方数据