A C Compiler for Large Data Sequential Processing using Remote Memory
A C Compiler for Large Data Sequential Processing using Remote Memory
复制标题
使用远程存储器进行大数据顺序处理的 C 编译器
DOI：
10.1109/pacrim.2011.6032892
复制
发表时间：
2011
期刊：
影响因子：
--
通讯作者：
H. Midorikawa
中科院分区：
文献类型：
--
作者：
S. Yoshimura;H. Midorikawa
流行的64位操作系统使我们能够在计算机编程中使用大的内存地址空间。然而，实际的物理内存成为充分利用它的限制。当一个程序需要比计算机中可用的物理内存更多的内存时，传统的虚拟内存系统会在本地硬盘和物理内存之间执行页面交换。这里，随着高速网络的最近发展，经由网络的远程存储器访问变得比访问本地硬盘更快。我们构建了分布式大内存系统（DLM）来访问网络中的大量远程内存。DLM被设计为用户级软件，具有很高的可移植性。DLM使用分布在群集节点上的远程内存提供非常大的虚拟内存。本文提出了一种新设计的用于DLM的C编译器。它提供了一个简单的编程接口，可以将DLM的丰富内存用于现有的顺序程序，而不是开发并行程序。
Prevailing 64bit-OS enables us to use a large memory address space in computer programming general. However, the actual physical memory becomes the limitation in utilizing it fully. When a program requires more memory than available physical memory in a computer, a traditional virtual memory system performs the page swap between a local hard disk and physical memory. Here, with the recent development in high-speed network, remote-memory access via networks becomes faster than accessing a local hard disk. We built the Distributed Large Memory System (DLM) to access vast remote memories in networks. The DLM is designed as a user-level software for high portability. The DLM provides a very large virtual memory using remote memories distributed over cluster nodes. This paper proposes a newly designed C compiler for the DLM. It provides an easy programming interface to use the abundant memory of the DLM with existing sequential programs, instead of developing parallel programs.