Optimization Algorithm was developed for the simula ti on of ceramic grain growth at atomistic scale. Based on the coordination informa tion of different atoms, a structure of trident tree was applied to save large q ...Optimization Algorithm was developed for the simula ti on of ceramic grain growth at atomistic scale. Based on the coordination informa tion of different atoms, a structure of trident tree was applied to save large q uantities data, so as to solve the problems of large data information and long r unning time. For every atom a binary tree was firstly formed according to the X coordination of atom. If the values of X coordination were the same, the middle sub-tree of first layer formed then a binary tree according to the Y coordinati on of atom. If the values of Y coordination were also the same, the middle sub- tree of second layer formed then a binary tree according to the Z coordination o f atom. In this way the speed of whole program is enhanced obviously. In order t o reduce memory, in this structure only need to store the exterior atoms’ infor mation, an integer is used to store the interior atoms’ information. If other a toms take up an atom’s all adjacent positions, this atom will be deleted in the data structure, for all the adjacent positions’ atoms, the integer’s relative bit will be set 1 to denote that there is an atom in this position but not be s tored in the trident tree. When an outside atom is deleted, for all the bits tha t are set 1,an atom will be added to the trident tree as an outside atom for the relative positions. And for this new added atom, the integer’s relative bi t of all the adjacent position’s atoms should be set 0 to denote that there is no interior atom in this position. In this way, if there are n 3 atoms, onl y need to store 6n 2 quantity’s atoms’ information. Large quantity of mem ory space can then be saved.展开更多
针对最大互信息系数(maximal information coefficient,MIC)近似算法在大规模数据场景下的计算时间复杂度高,计算时间增长快的问题,提出一种最大互信息系数并行计算(parallel computing maximal information coefficient,PCMIC)方法。...针对最大互信息系数(maximal information coefficient,MIC)近似算法在大规模数据场景下的计算时间复杂度高,计算时间增长快的问题,提出一种最大互信息系数并行计算(parallel computing maximal information coefficient,PCMIC)方法。分别在Spark和Spark-MPI(message passing interface)计算框架中,在不同的数据规模和不同的噪声水平下,利用PCMIC算法对14种典型的相关关系做并行计算。另外在不同节点数的情况下,选择两种具有代表性的相关关系来测试PCMIC算法在两种计算框架中的性能。结果表明:PCMIC算法在两种框架下的运算效果与原始MIC近似算法相比,同样具有普适性和均匀性,而且具有良好的可扩展性;随着数据规模和节点数的增加,PCMIC算法在两种框架中运算的时间增长明显比MIC近似算法慢,而且在Spark-MPI框架下的并行加速比和效率略优于Spark;Spark能够支持MPI任务的调度,为研究不同并行计算框架之间的融合奠定了一定的理论和应用基础。展开更多
文摘Optimization Algorithm was developed for the simula ti on of ceramic grain growth at atomistic scale. Based on the coordination informa tion of different atoms, a structure of trident tree was applied to save large q uantities data, so as to solve the problems of large data information and long r unning time. For every atom a binary tree was firstly formed according to the X coordination of atom. If the values of X coordination were the same, the middle sub-tree of first layer formed then a binary tree according to the Y coordinati on of atom. If the values of Y coordination were also the same, the middle sub- tree of second layer formed then a binary tree according to the Z coordination o f atom. In this way the speed of whole program is enhanced obviously. In order t o reduce memory, in this structure only need to store the exterior atoms’ infor mation, an integer is used to store the interior atoms’ information. If other a toms take up an atom’s all adjacent positions, this atom will be deleted in the data structure, for all the adjacent positions’ atoms, the integer’s relative bit will be set 1 to denote that there is an atom in this position but not be s tored in the trident tree. When an outside atom is deleted, for all the bits tha t are set 1,an atom will be added to the trident tree as an outside atom for the relative positions. And for this new added atom, the integer’s relative bi t of all the adjacent position’s atoms should be set 0 to denote that there is no interior atom in this position. In this way, if there are n 3 atoms, onl y need to store 6n 2 quantity’s atoms’ information. Large quantity of mem ory space can then be saved.
文摘针对最大互信息系数(maximal information coefficient,MIC)近似算法在大规模数据场景下的计算时间复杂度高,计算时间增长快的问题,提出一种最大互信息系数并行计算(parallel computing maximal information coefficient,PCMIC)方法。分别在Spark和Spark-MPI(message passing interface)计算框架中,在不同的数据规模和不同的噪声水平下,利用PCMIC算法对14种典型的相关关系做并行计算。另外在不同节点数的情况下,选择两种具有代表性的相关关系来测试PCMIC算法在两种计算框架中的性能。结果表明:PCMIC算法在两种框架下的运算效果与原始MIC近似算法相比,同样具有普适性和均匀性,而且具有良好的可扩展性;随着数据规模和节点数的增加,PCMIC算法在两种框架中运算的时间增长明显比MIC近似算法慢,而且在Spark-MPI框架下的并行加速比和效率略优于Spark;Spark能够支持MPI任务的调度,为研究不同并行计算框架之间的融合奠定了一定的理论和应用基础。
