分层检查点的近似最优周期计算模型

发布时间：2018-01-28 04:47

本文关键词： 高性能计算容错分层检查点检查点周期近似最优解　出处：《计算机应用》2017年01期 　论文类型：期刊论文

【摘要】：针对大规模高性能计算(HPC)系统中检查点效率提升问题,提出一种面向分层检查点近似最优周期计算模型。首先,通过分析一个HPC系统中应用程序的执行过程,将检查点周期优化抽象为一个非线性的检查点成本模型;其次,通过分析可能故障位置推导出分层检查点成本公式,并引入两个减速因子和一个加速因子来模拟消息日志对分层检查点造成的影响。仿真实验结果表明,所提模型与理论近似最优周期检查点成本平均误差在5%以下,相对传统检查点周期优化模型的平均误差降低了20%,能够有效提高检查点的效率,提升HPC系统可用性。
[Abstract]:Aiming at the problem of improving checkpoint efficiency in large scale high performance computing (HPC) systems, an approximate optimal period calculation model for layered checkpoint is proposed. By analyzing the execution process of an application program in a HPC system, the checkpointing cycle optimization is abstracted into a nonlinear checkpoint cost model. Secondly, the cost formula of hierarchical checkpoint is deduced by analyzing the possible fault location, and two deceleration factors and one acceleration factor are introduced to simulate the effect of message log on hierarchical checkpoint. The average cost error of the proposed model is below 5%, compared with the traditional checkpointing cycle optimization model, the average error is reduced by 20%, which can effectively improve the efficiency of checkpoint. Improve HPC system availability.
【作者单位】：哈尔滨工程大学计算机科学与技术学院;
【基金】：国家自然科学基金资助项目(61370212,61402127,61502118) 黑龙江省自然科学基金资助项目(F2015029)~~
【分类号】：TP38
【正文快照】： 0引言随着大规模和超大规模集成电路的问世,高性能计算(High Performance Computation,HPC)系统进入高速发展期,根据International Exascale Software Project(IESP)的研究报告[1]显示,HPC系统及其相关技术会持续发展。然而在实际部署与运行中人们发现,系统的高复杂性、高异

【相似文献】