-
公开(公告)号:US20130024871A1
公开(公告)日:2013-01-24
申请号:US13545262
申请日:2012-07-10
申请人: Yaoqing Gao , Francois Gibello , Eric Michel , Nicolas Tallet , Pascal Vezolle
发明人: Yaoqing Gao , Francois Gibello , Eric Michel , Nicolas Tallet , Pascal Vezolle
IPC分类号: G06F9/46
CPC分类号: G06F9/5083 , G06F9/485
摘要: A method and system are provided for thread management in parallel processes in a multi-core or multi-node system. The method includes receiving monitored hardware metrics information from the multiple cores or multiple nodes on which processes are executed, receiving monitored process and thread information; and globally monitoring the processing across the multiple cores or multiple nodes. The method further includes analyzing the monitored information to minimize imbalances between the multiple cores and/or to improve core or node exploitation and dynamically adjusting the number of threads per process based on the analysis.
摘要翻译: 提供了一种用于多核或多节点系统中并行进程中的线程管理的方法和系统。 该方法包括从执行进程的多个核心或多个节点接收受监视的硬件度量信息,接收被监控的进程和线程信息; 并全面监控跨多个核心或多个节点的处理。 该方法还包括分析所监视的信息以最小化多个核之间的不平衡和/或改善核心或节点利用并基于分析动态地调整每个进程的线程数。
-
公开(公告)号:US08990830B2
公开(公告)日:2015-03-24
申请号:US13545262
申请日:2012-07-10
申请人: Yaoqing Gao , Francois Gibello , Eric Michel , Nicolas Tallet , Pascal Vezolle
发明人: Yaoqing Gao , Francois Gibello , Eric Michel , Nicolas Tallet , Pascal Vezolle
CPC分类号: G06F9/5083 , G06F9/485
摘要: A method and system are provided for thread management in parallel processes in a multi-core or multi-node system. The method includes receiving monitored hardware metrics information from the multiple cores or multiple nodes on which processes are executed, receiving monitored process and thread information; and globally monitoring the processing across the multiple cores or multiple nodes. The method further includes analyzing the monitored information to minimize imbalances between the multiple cores and/or to improve core or node exploitation and dynamically adjusting the number of threads per process based on the analysis.
摘要翻译: 提供了一种用于多核或多节点系统中并行进程中的线程管理的方法和系统。 该方法包括从执行进程的多个核心或多个节点接收受监视的硬件度量信息,接收被监控的进程和线程信息; 并全面监控跨多个核心或多个节点的处理。 该方法还包括分析所监视的信息以最小化多个核之间的不平衡和/或改善核心或节点利用并基于分析动态地调整每个进程的线程数。
-