Abstract:
This paper analyzes workload characteristics and power usage characteristics of parallel high-performance computing system, proposes a power management algorithm to reduce the power consumption of these systems by restricting the resource allocation and controlling the power state(sleep or active) of nodes. Detailed experiment using data from Parallel Workloads Archive tests the power management algorithm. Experimental result shows that the algorithm can achieve considerable overall system energy savings with small or no performance impact.
Key words:
high-performance computing,
power management,
scheduling,
resource allocation,
workload
摘要: 分析并行高性能计算系统的负载及能耗特征,提出一种功耗管理算法,通过控制节点的分配和功耗状态(睡眠状态或激活状态)降低系统的能耗,并使用Parallel Workloads Archive的负载数据对功耗管理算法进行测试。实验结果表明,该算法在不影响或较少牺牲系统性能的情况下,能有效节省系统能耗。
关键词:
高性能计算,
功耗管理,
调度,
资源分配,
作业负载
CLC Number:
DONG Jing, LEI Yu-Tong, CHEN Hai-Chao. Power Management of High-Performance Computing System Based on Resource Limiting[J]. Computer Engineering, 2010, 36(16): 276-277.
董晶, 卢宇彤, 陈海涛. 基于资源限制的高性能计算系统功耗管理[J]. 计算机工程, 2010, 36(16): 276-277.