Pseudospectral method is an efficient and high accuracy numerical method for simulating seismic wave propaga- tion in heterogeneous earth medium. Since its derivative operator is global, this method is commonly consid...Pseudospectral method is an efficient and high accuracy numerical method for simulating seismic wave propaga- tion in heterogeneous earth medium. Since its derivative operator is global, this method is commonly considered not suitable for parallel computation. In this paper, we introduce the parallel overlap domain decomposition scheme and give a parallel pseudospectral method implemented on distributed memory PC cluster system for modeling seismic wave propagation in heterogeneous medium. In this parallel method, the medium is decomposed into several subdomains and the wave equations are solved in each subdomain simultaneously. The solutions in each subdomain are connected through the transferring at the overlapped region. Using 2D models, we compared the parallel and traditional pseudospectral method, analyzed the accuracy of the parallel method. The results show that the parallel method can efficiently reduce computation time for the same accuracy as the traditional method. This method could be applied to large scale modeling of seismic wave propagation in 3D heterogeneous medium.展开更多
提出一种新的面向复杂网络大数据的重叠社区检测算法DOC(detecting overlapping communities over complex network big data),时间复杂度为O(nlog2(n)),算法基于模块度聚类和图计算思想,应用新的节点和边的更新方法,利用平衡二叉树对...提出一种新的面向复杂网络大数据的重叠社区检测算法DOC(detecting overlapping communities over complex network big data),时间复杂度为O(nlog2(n)),算法基于模块度聚类和图计算思想,应用新的节点和边的更新方法,利用平衡二叉树对模块度增量建立索引,基于模块度最优的思想设计一种新的重叠社区检测算法.相对于传统的重叠节点检测算法,对每个节点分析的频率大为降低,可以在较低的算法运行时间下获得较高的识别准确率.复杂网络大数据集上的算法测试结果表明:DOC算法能够有效地检测出网络重叠社区,社区识别准确率较高,在大规模LFR基准数据集上其重叠社区检测标准化互信息指标NMI最高能达到0.97,重叠节点检测指标F-score的平均值在0.91以上,且复杂网络大数据下的运行时间明显优于传统算法.展开更多
将互动直播部署在边缘计算环境中,可以在网络边缘对直播视频进行转码和传输,通过用户附近的边缘服务器提供低延迟的直播服务.然而,在多边缘服务器、多用户场景下存在着直播用户分配问题,导致直播用户体验质量(quality of experience, Q...将互动直播部署在边缘计算环境中,可以在网络边缘对直播视频进行转码和传输,通过用户附近的边缘服务器提供低延迟的直播服务.然而,在多边缘服务器、多用户场景下存在着直播用户分配问题,导致直播用户体验质量(quality of experience, QoE)无法得到保证.为了提高直播用户QoE,需要根据用户的个性化需求合理地分配服务器资源.首先分析真实数据集,发现大多数用户处于多基站重叠覆盖区域内,并且不同用户的互动需求存在差异;然后根据互动直播的特点提出一种适用于边缘计算场景的用户QoE模型,该模型综合考虑了直播用户的视频质量和互动体验;最后设计一种高效的直播用户分配算法,优化了多边缘服务器重叠覆盖区域内的直播用户QoE.仿真实验表明,所提出的用户分配策略可为用户提供高码率和低延迟的直播视频,同时能有效降低边缘服务器切换次数和码率抖动,使直播用户QoE相较于其他策略提升超过19%.展开更多
采用Mortar有限单元法(mortar finite element method,MFEM)能够得到正定、对称的系数矩阵,而且刚度矩阵是分块对称的,这种特点适合于并行迭代求解。阐述了非重叠Mortar有限单元法(non-overlapping MFEM,NO-MFEM)的基本原理,介绍了适合...采用Mortar有限单元法(mortar finite element method,MFEM)能够得到正定、对称的系数矩阵,而且刚度矩阵是分块对称的,这种特点适合于并行迭代求解。阐述了非重叠Mortar有限单元法(non-overlapping MFEM,NO-MFEM)的基本原理,介绍了适合于NO-MFEM并行计算的区域分解策略以及并行求解的基本流程。针对简单2维静电场问题,使用NO-MFEM进行了并行计算,并与理论值和串行计算结果进行对比,验证了所提方法的有效性。同时,对于非协调网格造成的计算误差进行了分析。NO-MFEM法的并行计算为工程应用中优化设计问题的区域分解和并行求解提供了一种新的选择。展开更多
基金National Natural Science Foundation of China (40474012 and 40521002)
文摘Pseudospectral method is an efficient and high accuracy numerical method for simulating seismic wave propaga- tion in heterogeneous earth medium. Since its derivative operator is global, this method is commonly considered not suitable for parallel computation. In this paper, we introduce the parallel overlap domain decomposition scheme and give a parallel pseudospectral method implemented on distributed memory PC cluster system for modeling seismic wave propagation in heterogeneous medium. In this parallel method, the medium is decomposed into several subdomains and the wave equations are solved in each subdomain simultaneously. The solutions in each subdomain are connected through the transferring at the overlapped region. Using 2D models, we compared the parallel and traditional pseudospectral method, analyzed the accuracy of the parallel method. The results show that the parallel method can efficiently reduce computation time for the same accuracy as the traditional method. This method could be applied to large scale modeling of seismic wave propagation in 3D heterogeneous medium.
文摘提出一种新的面向复杂网络大数据的重叠社区检测算法DOC(detecting overlapping communities over complex network big data),时间复杂度为O(nlog2(n)),算法基于模块度聚类和图计算思想,应用新的节点和边的更新方法,利用平衡二叉树对模块度增量建立索引,基于模块度最优的思想设计一种新的重叠社区检测算法.相对于传统的重叠节点检测算法,对每个节点分析的频率大为降低,可以在较低的算法运行时间下获得较高的识别准确率.复杂网络大数据集上的算法测试结果表明:DOC算法能够有效地检测出网络重叠社区,社区识别准确率较高,在大规模LFR基准数据集上其重叠社区检测标准化互信息指标NMI最高能达到0.97,重叠节点检测指标F-score的平均值在0.91以上,且复杂网络大数据下的运行时间明显优于传统算法.
文摘将互动直播部署在边缘计算环境中,可以在网络边缘对直播视频进行转码和传输,通过用户附近的边缘服务器提供低延迟的直播服务.然而,在多边缘服务器、多用户场景下存在着直播用户分配问题,导致直播用户体验质量(quality of experience, QoE)无法得到保证.为了提高直播用户QoE,需要根据用户的个性化需求合理地分配服务器资源.首先分析真实数据集,发现大多数用户处于多基站重叠覆盖区域内,并且不同用户的互动需求存在差异;然后根据互动直播的特点提出一种适用于边缘计算场景的用户QoE模型,该模型综合考虑了直播用户的视频质量和互动体验;最后设计一种高效的直播用户分配算法,优化了多边缘服务器重叠覆盖区域内的直播用户QoE.仿真实验表明,所提出的用户分配策略可为用户提供高码率和低延迟的直播视频,同时能有效降低边缘服务器切换次数和码率抖动,使直播用户QoE相较于其他策略提升超过19%.
文摘采用Mortar有限单元法(mortar finite element method,MFEM)能够得到正定、对称的系数矩阵,而且刚度矩阵是分块对称的,这种特点适合于并行迭代求解。阐述了非重叠Mortar有限单元法(non-overlapping MFEM,NO-MFEM)的基本原理,介绍了适合于NO-MFEM并行计算的区域分解策略以及并行求解的基本流程。针对简单2维静电场问题,使用NO-MFEM进行了并行计算,并与理论值和串行计算结果进行对比,验证了所提方法的有效性。同时,对于非协调网格造成的计算误差进行了分析。NO-MFEM法的并行计算为工程应用中优化设计问题的区域分解和并行求解提供了一种新的选择。
