I/O受限的并行加速比模型与可扩展I/O体系结构

An I/O Restricted Parallel Speedup Model and the Scalable I/O Architecture

为了缓解I/O瓶颈问题,可以从应用程序、可扩展算法、编译器和语言、运行时库、操作系统和体系结构六方面展开研究。其中,I/O体系结构是所有技术途径的关键支撑。当前并行I/O性能分析缺乏科学的理论模型为I/O体系结构设计提供理论依据。本文针对并行计算机系统的可扩展性问题,研究了I/O负载对并行计算机系统可扩展性的影响,建立了I/O受限的并行加速比性能模型,对目前大规模并行计算机系统中三种常用I/O体系结构的可扩展性进行了分析;以此为理论依据,提出了一种面向高性能计算的可扩展并行I/O系统结构。同时,还提出了几种有效降低I/O操作服务时间的策略,从而达到增强系统可扩展性的目的,为后续研究奠定了基础。

The effective solutions for the I/O bottleneck can be found from the following six levels, including applications, algorithms, languages and compilers, run-time libraries, operating systems, and I/O architecture. Among all the levels mentioned above, the I/O architecture is the most fundamental. In order to meet the I/O requirements and challenges, along with our research task of a high performance parallel computing system, this paper presents a theoretical study of I/O architectures, from which we can make it possible the high performance and scalability in terms of I/O architecture level. The current parallel I/O performance analysis lacks scientific theoretical models to support the I/O architecture design. The paper studies the impact of I/O workload on the scalability of parallel computing systems and proposes an I/O restricted parallel speedup model. Based on this model, which can be used to guide I/O architecture design, a scalable parallel I/O architecture for high performance computing is presented. Moreover, the paper analyzes several strategies for improving the system scalability, which serves as the basis for further study.