面向大数据的高时效并行计算机系统结构与技术

Currently,Big Data processing in different areas,such as commercial intelligence, social management and scientific computing, has requirements ofhigh performance and low latency, leading to serious challenges for time-efficient computing in traditional data processing infrastructures. This raises urgent needs of significant renovations in the data access methods, processing hardware and software supporting for Big Data. Accordingly, we propose to research on the architecture and key technology of high performance and time-efficient parallel computer for Big Data. In the aspect of parallel computer architecture, we focus on the high performance accelerator architecture and the scalable heterogeneous parallel computing system architecture. In the aspect of hybrid main memory, we will mainly explore the streaming and high concurrent main memory structures and the optimization mechanisms for the data management of hybrid main memory. In the aspect of parallel processing software supporting, we will mainly study the heterogeneous platform-based task-granularity partitioning and efficient planning mechanism as well as the data layouts mechanisms targeting high throughput and low latency. At last, we will evaluate and verify our research results through experiments over Big Data collections from real-world applications. This project has both significant academic value and application prospect. The research results will provide fundamental theory, methods and techniques for designing and developing high performance and time-efficient computer systems for Big Data analysis and processing..

当前，商业智能、社会管理、科学计算等多个领域的大数据处理应用都蕴含了高性能、低延迟的需求，对传统数据处理设施带来了高时效计算挑战，迫切需要在大数据的存取方法、处理硬件和软件支撑方面作出重大革新。为此，提出了面向大数据的高时效并行计算机系统结构与技术的研究课题。在并行计算系统体系结构方面，主要研究面向大数据的高性能加速处理器结构和大数据的可扩展异构并行计算体系结构；在混合主存系统方面，主要研究面向大数据的流式高并发主存结构和混合主存的数据管理优化机制；在并行处理的软件支撑方面，主要研究基于异构平台的任务粒度划分与高效规划机制和并行处理系统中高吞吐、低延迟的数据布局机制。最后，通过实际应用的大数据进行测试验证研究成果。本项目研究具有重要的学术价值和应用前景，为设计和研制高时效的大数据分析和处理计算系统提供了基础理论方法和技术。

当前，商业智能、社会管理、科学计算等多个领域的大数据处理应用都蕴含了高性能、低延迟的需求，对传统数据处理设施带来了高时效计算挑战，迫切需要在大数据的存取方法、处理硬件和软件支撑方面作出重大革新。为此，提出了面向大数据的高时效并行计算机系统结构与技术的研究课题。在并行计算系统体系结构方面，主要研究面向大数据的高性能加速处理器结构和大数据的可扩展异构并行计算体系结构；在混合主存系统方面，主要研究面向大数据的流式高并发主存结构和混合主存的数据管理优化机制；在并行处理的软件支撑方面，主要研究基于异构平台的任务粒度划分与高效规划机制和并行处理系统中高吞吐、低延迟的数据布局机制。最后，通过实际应用的大数据进行测试验证研究成果。本项目研究具有重要的学术价值和应用前景，为设计和研制高时效的大数据分析和处理计算系统提供了基础理论方法和技术。