面向大规模集群并行I/O的用户层配置优化策略研究

近年来,大规模并行场景下的I/O性能越来越受到应用科学家的关注。影响应用I/O性能的关键因素主要有三个层次：包括应用的I/O接口实现、体系结构和文件系统组件的性能以及应用的I/O参数配置。本文从应用I/O配置优化的视角,分析了大规模集群并行I/O的配置调优空间,在此基础上,给出了一套大规模集群并行I/O性能特征测试分析方法,基于该方法,在某国产超级计算集群上开展了一系列I/O测试分析来刻画系统的I/O性能特征,进而指导并行应用程序的I/O配置优化。基于优化后的配置参数,在两类典型的并行I/O场景中,针对某类生产应用程序,8192进程下的重启动数据写操作时间下降了15%,4096核的程序作业加载时间从10分钟缩短到了5s。本文提出的I/O配置调优空间及优化方法,可以推广应用到其它同类系统平台,对于大规模集群上的用户层并行I/O配置调优具有借鉴意义。     

面向大规模集群的并行I/O用户层配置优化策略

影响应用I/O性能的关键因素主要有三个层次:包括应用的I/O接口实现、体系结构和文件系统组件的性能以及应用的I/O参数配置。从应用I/O配置优化的视角,分析了大规模集群并行I/O的配置调优空间,在此基础上,给出了一套大规模集群并行I/O性能特征测试分析方法。基于该方法,在某国产超级计算集群上开展了一系列I/O测试分析来刻画系统的I/O性能特征,进而指导并行应用程序的I/O配置优化。基于优化后的配置参数,在两类典型的并行I/O场景中,针对某类生产应用程序,8192进程下的重启动数据写操作时间下降了15%,4096核的程序作业加载时间从10 min缩短到了5 s。     

导弹对抗离散数学模型

针对现代非接触战争的特点,用数学方法研究导弹对抗模型问题,针对作战的不同目的和要求,建立两个导弹对抗的非线性规划数学模型,对模型求解设计了近似方法,该方法把上述非线性规划模型化为多个线性规划模型进行求解.对一类较简单的特殊情况进行了仔细分析,得到了分析解.最后指出该方法如何应用于动态实时优化决策.本文的方法可用来预测和评估导弹对抗的效果,为作战实时优化决策提供参考.     

远程预警声呐发现目标后的机动原则

鉴于经典的载体优化理论对装备舷侧阵声呐的潜艇的局限性,讨论了装备舷侧阵远程预警声呐的潜艇发现目标后的机动问题.在分析经典的载体优化机动理论的使用条件和特点的前提下,借鉴美国潜艇拖曳阵远程预警声呐发现目标后的战术机动思想,提出了装备舷侧阵声呐的潜艇发现目标后可以采取与初始发现方位线相垂直的倒"U"字机动策略及具体的细节.该机动方法理论上成立,但具体参数的确定还需要经过计算机仿真和海上实践的检验.同时认为战术理论的发展与装备技术水平的发展紧密相关.     

Optimality of the greedy shooting strategy in the presence of incomplete damage information

Consider a situation where a single shooter engages, sequentially, a cluster of targets that may vary in terms of vulnerability and value or worth. Following the shooting of a round of fire at a certain target, the latter may either be killed or remain alive. We assume neither partial nor cumulative damage. If the target is killed, there is a possibility that the shooter is not aware of that fact and may keep on engaging that target. If the shooter recognizes a killed target as such, then this target is considered to be evidently killed. If the objective is to maximize the weighted expected number of killed targets, where the weight reflects the value of a target, then it is shown that a certain type of a shooting strategy, called a Greedy Strategy, is optimal under the general assumption that the more a target is engaged, but still not evidently killed, the less is the probability that the next round will be effective. If all weights are equal, then the greedy shooting strategy calls to engage, at each round, the least previously engaged target that is not evidently killed. © 1997 John Wiley & Sons, Inc. Naval Research Logistics 44: 613–622, 1997     

‘Mad Dog?’ Samuel Huntington and the Vietnam War

Harvard professor Samuel P. Huntington has frequently been considered a Vietnam War hawk. His observation that ‘forced-draft urbanization’ might help the United States win the war has come to define his engagement in contemporary strategic debates. This essay argues that both Huntington’s academic work and his private policy advice to the U.S. Government in fact urged a political settlement to the conflict. It argues that in spite of this, Huntington refused to break publicly with the U.S. policy because of his wider concern over what he saw as a crisis of authority in the U.S. foreign policy and governing institutions in the era.     

Small footprint,small payoff: The military effectiveness of security force assistance

After 15 years of war in Afghanistan and Iraq, many now see ‘small-footprint’ security force assistance (SFA) – training, advising and equipping allied militaries – as an alternative to large US ground-force commitments. Yet, its actual military efficacy has been little studied. This paper seeks to fill this gap. We find important limitations on SFA’s military utility, stemming from agency problems arising from systematic interest misalignment between the US and its typical partners. SFA’s achievable upper bound is modest and attainable only if US policy is intrusive and conditional, which it rarely is. For SFA, small footprints will usually mean small payoffs.     

A nation-in-the-making,in arms: control of force,strategy and the Ukrainian Volunteer Battalions

At the core of waging war and strategy is the creation, control and use of force. This article investigates the volunteer battalions that mobilized in Ukraine during the spring of 2014. It contextualizes the volunteer phenomenon and focuses on the state strategies to establish control over these militias. As ambiguous entities arising from a situation characterized by rapid social change – revolution and war – the volunteer battalions threatened existing hierarchies and questioned state authority. The situation was exacerbated by the war, which deviated from the expectations of Ukrainian combatants and Western military observers alike. The state nevertheless enjoyed a modicum of success in reining in the militias through four strategies of undermining, co-option, incorporation and coercion. While predominantly integrated into a more rigid category of paramilitary forces, the volunteers continue to play a role in both the Ukrainian society and security sector to the unforeseeable future.     

美军战略传播中的“叙事”策略

随着互联网技术的飞速发展,信息传播对国家政治、经济、外交和国防带来巨大冲击和影响。美军意识到传播叙事是政府和军队塑造形象、谋划战略、输出文化价值观的重要手段。美国从"重讲故事"的角度出发,将"叙事"包装成对外战略传播的新型方式,从思维认知、情感操控、舆论引导等方面强化对对手的心理威慑和影响,同时以新媒体为媒介扩大叙事传播的影响受众,进而达到物理打击所难以发挥的影响。     

雾霭图像实时处理算法加速策略

为仿真前视红外导引头实时匹配导引过程,完成目标实时跟踪性判断,针对限制对比度自适应直方图均衡和归一化互相关算法开展加速策略研究,提出基于统一计算设备架构(Compute Unified Device Architecture, CUDA)的雾霭图像实时处理加速方案,并进行设计参数寻优,得出线程块、线程网格设计参数存在最优尺寸,需实验测定的结论。实验结果表明,引入CUDA加速优化后,较未进行CUDA加速前时间指标提高5~20倍,能够满足目标跟踪实时性要求,可为红外导引头目标实时跟踪系统设计提供参考。