【作者】 阚博文； 石京燕；

【Author】 Bowen Kan, Jingyan Shi (Institute of High Energy Physics, Beijing 100049)

【机构】 中国科学院高能物理研究所计算中心；

【摘要】 Torque[1]衍生于PBS,是一个高效的开源计算资源管理软件,最早由NASA的Ames研究中心开发,用于满足异构网络的计算需求。随着分布式计算技术的发展,Torque在高性能计算集群中得到广泛应用。但是作为开源软件,目前缺乏一个相对完善的监控系统,监视torque运行的实时状态,记录其历史运行情况,并实现部分控制功能,保证Torque运行稳定性、可靠性并提高其易管理性。为此,本文从以下五个方面设计实现了针对Torque的自适应轻量级监控系统:1)实现一个轻量级日志循环过滤功能,实时获取Torque的作业运行状态;2)为管理员提供可自定义的监控命令统一接口,用于查询Torque管理的的各类资源信息;3)设计了监控信息存储策略,使监控信息持久化;4)为用户提供了可自定义的报警接口,能够将Torque时运行出现的异常错误以邮件或短信的方式实时提交给用户;5)利用HTML5技术,实现了Torque管理的作业状态可视化,实时化,并提供管理员Torque运行状态展示内容的可定制功能。更多还原

【Abstract】 Torque, an efficient PBS(Portable Batch System)-based open source Resource Management system, was originally developed by Ames research center of NASA, which was designed to satisfy the computing requirements of heterogeneous network. With the development of distributed computing, Torque has been widely used in high performance computing cluster. However, because of the lack of a well designed monitoring system, it is difficult to monitor, record, and control , leading to low stability, reliability and manageability. To overcome those problems, this paper designs and implements an adaptive lightweight monitoring system for torque from five aspects. 1) A lightweight circulating filtration logging system is developed to obtain the real-time running status of torque; 2) One uniform interface was provided for administrators to define monitoring commands, which can query management resources of torque; 3) Storage strategy is designed to make monitoring information persistent; 4) One uniform interface is provided for users to customized alarms, which can submit exceptions and errors to users via emails and SMS in real time; 5) HTML5 technology is applied in the customizable visualization of the jobs’ status in torque in real time.更多还原