| 摘要: |
| 中国科学院超级计算环境是整合了包括总中心、分中心和所级中心计算资源的3层架构超级计算环境.为提升超级计算环境的可靠性，提供稳定、可靠的计算服务，其容错机制的研究成为超级计算环境的一个研究重点.在对容错基本思想及各类计算机容错技术进行充分调研的基础上，提出一种适用于超级计算环境的容错框架，依据该框架给出了不同层次的容错方案，并对不同层次的容错开销进行了分析和比较，验证了不同层次容错方案对应用程序所带来的影响. |
| 关键词: 超级计算 容错框架 检查点设置/回卷恢复 容错开销 |
| DOI： |
| 分类号: |
| 基金项目:国家高技术研究发展计划（863）（2011AA01A205）；中国科学院知识创新工程青年人才领域项目（CNIC_QN_10004）；中国科学院青年创新促进会基金 |
|
| Fault-Tolerant Mechanism in Supercomputing Environment |
|
ZHAO Yi, CAO Zong-Yan, ZHU Peng, CHI Xue-Bin
|
|
Supercomputing Center, Computer Network Information Center, The Chinese Academy of Sciences, Beijing 100190, China
|
| Abstract: |
| The three layers supercomputing environment of Chinese Academy of Sciences is built to integrate the computing resources of the head center in Beijing, eight regional centers and several campus-level centers. To enhance the reliability of the supercomputing environment and provide stable and reliable computing services, the fault-tolerant mechanism research has become a research priority of the supercomputing environment. In this paper, the fault-tolerant basic concepts and computer fault-tolerant technologies are introduced at first. Next, a fault-tolerant framework of the supercomputing environment is proposed. Then the fault-tolerant solutions of different levels based on the framework and the performance test results in Deepcomp 7000 are presented. Finally, the fault-tolerant overheads of different levels are compared and analyzed to verify the impact on the application. |
| Key words: supercomputing fault-tolerant framework checkpoint/rollback recovery fault-tolerant overheads |
