It can lighten programmers ’ burden . the performance analysis and evaluation module analyzes speedup , efficiency and scalability and give the computation formula . this paper also brings forward a performance analysis celibrating method and a practical method to predict program execution time and communication time and analyze performance of parallel programs 性能分析与评测模块对加速比、效率、可扩展性等进行分析,给出了它们的计算公式,以及性能分析的量化方法和实用的性能评价方法,预测程序通信、执行时间,分析程序性能。
We summarized several main reasons of low parallel efficiency of the npupar system , and improved it accordingly . in our test , the efficiency of the programs generated by modified npupar increased remarkablely ( about 10 % ) . we also make deep quantitative analysis for the super - linear speedup in the test 通过对两个实际的cfd实例程序的对比测试，改进后的npupar系统在保证正确性的基础上并行效率有了明显的提高(普遍提高了10左右) ，并对测试中出现的超线性加速比做了深入的定量分析。
The discussion of main parallel technologies on construction of parallel sliq algorithm is presented in this paper . the computing result of algorithm complexity of sequential and parallel algorithm indicates : when the data set is large enough , as to continuous attributes , the parallel algorithm almost get speedup value equal to the number of processors ， while as to categorical attribute the improvement of parallel algorithm is limited 通过对串行和并行算法时间复杂度的计算表明，当数据集充分大时，由于连续属性的排序计算操作分散到各个处理机单元上进行，显著降低了计算时间，从而可以得到近似于处理机个数的加速比，对于离散属性，本并行算法对串行算法的性能提高有限
This divide - and - conquer algorithm is combined with multisection method when it ' s implemented under distributed environment , high speedup and efficiency can be achieved , ii ) a parallel divide - and - conquer algorithm combined with multisection method is proposed , the sum of the subproblem ' s scales is equal to the original problem ' s scale minus r , where r is semi - bandwidth 在并行实现时，算法结合并行性好的多分法，在分布式环境下获得了很高的并行加速比和效率。 ii )提出了另一种结合多分法的并行分治算法，给出了特征值分割定理及其证明。在该算法中，子问题的规模之和等于原问题的规模减; ，其中;为半带宽。
Followed by the rapid extension of data size , the usage of parallel technology is a very important method to improve the efficiency of data ming . sliq uses novel pre - sorting and breadth - first techniques to build a decision tree fast and accurately on a large data set , and can deal both categorical and numeric attributes . but the primary algorithm contains the abundant computing on attribute and record 本文首先分析了串行sliq算法的原理和特点，针对其不足提出了一些改进方法，然后在基于pvm的环境下实现了算法的并行化，分析了算法的时间复杂度和加速比，提高了sliq算法的效率，具有一定的理论意义和实用价值。
The entire design process and all of the key technologies of the ppdbm ( parallel programme dynamic behavior monitor ) are illustrated in the paper , a performance analysis and monitor tool based on the cluster system . it includes the implement of the software events recorder , the dynamic visualization of the status of the processes , the status of the cpus and memories of the running nodes , and the graphical show of the analyzed data of parallel accelerate rate , parallel efficiency and computing - communication ratio 论文给出了基于机群系统的并行程序动态行为监测系统( ppdbm )的设计过程，详细介绍了ppdbm的实现方案，其中主要包括软件事件记录器的实现，进程各事件行为状态、各节点cpu状态及内存信息动态可视化显示，以及并行加速比、并行效率、计算/通信比分析数据的图形化显示。
Also , the relactions between the best block size for matrix transpose and the size and associativity of the processor ' s cache is formulized . for parallel optimization , several programming models available on a numa system , such as lightweight processes ( sproc ) , posix threads , openmp and mpi , are compared , and their speedup and coding complexity are analyzed 对于sar成像处理的并行优化，本文对比了在numa架构上可用的几种并行编程模型：轻量级进程、 posix线程、 openmp和mpi ，针对numa架构和sar成像处理的特点从加速比、编程复杂度等多个方面进行了讨论。
In the thesis a new count expression of mesh fractal dimension is introduced , on the basis of it the storage and operation counts of directs solution methods , assembly global stiffness iterative methods and ebe iterative methods have been analyzed and compared , a conclusion has been obtained that the storage and operation counts of direct solution methods is the most , the storage counts of the ebe iterative methods is the least . estimating compute scale for ebe iterative methods at shuguang 2000 parallel computer , it is make out that compute scale is large in the extreme if make the most of the memory of every processor . by analyzing the parallel compute performance of 2d upend - step flow calculation , it is proved that ebe iterative methods can achieve upper speedup rations 本文还提出了一种网格分维新的计算公式，并基于网格分维对直接解法，组集总刚矩阵的迭代解法和基于ebe的迭代解法等三种方法的存储量和计算量进行了分析和比较，得出了直接解法无论存储量和计算量均最大，基于ebe迭代解法存储量最小的结论；在曙光2000并行机上估计了基于ebe迭代解法的计算规模，说明只要充分利用各处理机的内存，可解问题的规模是非常巨大的；对二维倒台阶流并行计算进行了性能分析，表明基于ebe策略的迭代解法能获得较高的加速比。
We focused our discussions on the mechanism of saturation of the wakefield and the electron parametric instabilities which affect the process of the wakefield generation and electrons acceleration . we developed a 2 - dimension distributed parallel pic code under mpi parallel environment and got a good speedup ratio tested on yh - iv and pc computer groups 为了解决研究激光和稀薄等离子体相互作用所需的大量计算，我们研制了21 2d分布式粒子模拟并行程序，在微机群和巨型机yh ? iv对程序并行性能进行了测试得到了较高的加速比。
Pvm , belongs to now , has been used widely . the paper implements the parallel algorithm of optimization bayesian classification on pvm , and analyzed acceleration rate and complexity . the analysis indicates that it is excellence when where is amount of class or the data is very large 本论文在pvm的基础上研究并实现了优化贝叶斯算法的并行化，并且分析了该算法的加速比和时间复杂度，分析表明在类比较多、或者待分类的数据样本比较多时，用该并行算法可以较大幅度提高数据分类的效率。
