当前位置:主页 > 科技论文 > 计算机论文 >

基于NBD机制的高性能计算机系统内存优化方案

发布时间:2018-09-04 10:51
【摘要】:随着多核、异构等计算机技术的飞速发展,高性能计算机的计算性能现已达到亿亿次量级.随着高性能计算机技术的成熟,高性能应用对系统内存产生不同的需求.超级计算机每个结点都按一定标准配置了内存,在实际使用中,因不同应用对于内存需求量的不同,很容易出现有的结点空闲内存较多,而有的结点内存紧张的情况.为较好地运行内存消耗量大的高性能应用,提出了一种基于网络块设备(network block device,NBD)机制使用远程结点内存构建page swap区的内存优化技术.该技术利用了超级计算机中通过网络访问远地内存性能优于访问本地磁盘性能的特点,采用了远程内存交换盘技术,使得本地节点在内存紧张时可将页面交换到远地节点的物理内存上.并将该技术应用到"天河一号"超级计算机系统上,以测试其性能并提高超级计算机的内存使用效率.
[Abstract]:With the rapid development of multi-core, heterogeneous computer technology, the computing performance of high performance computer has reached billions of times. With the maturity of high-performance computer technology, high-performance applications have different requirements for system memory. Each node of a supercomputer has allocated memory according to certain standards. In practice, due to the difference of memory requirements in different applications, it is easy to find some nodes have more free memory and some nodes have tight memory. In order to run high performance applications with large memory consumption, a memory optimization technique based on network block device (network block device,NBD) is proposed to construct page swap region using remote node memory. This technology takes advantage of the performance of accessing remote memory over network in supercomputers to access local disks, and adopts the technology of remote memory switch disk. Allows local nodes to swap pages to the physical memory of remote nodes when memory is tight. The technology is applied to the "Tianhe-1" supercomputer system to test its performance and improve the memory efficiency of the supercomputer.
【作者单位】: 国防科学技术大学计算机学院;
【分类号】:TP38;TP333


本文编号:2221864

资料下载
论文发表

本文链接:https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2221864.html


Copyright(c)文论论文网All Rights Reserved | 网站地图 |

版权申明:资料由用户71cd7***提供,本站仅收录摘要或目录,作者需要删除请E-mail邮箱bigeng88@qq.com