Parallel Arithmetic Researching For Simulation Based On LAN

The high-performance computational capacity has become a symbol to measure the high-tech development level of a country. The simulation based on the cluster of computers that utilizes the local area network technology as well as the real-time simulation of the continuous system on a large scale is an economical, convenient and effective way to solve the problems.On the basis of investigation on the latest literatures related with the parallel system software at home and abroad, this thesis has made some achievements:(1) Firstly, a partition model of medium granularity is initiated, which is adapted to the parallel simulation platform on the cluster of computers based on the local area network technology, and dependent on the modularized thought.The functions and practical algorithms of such modules as partitioning, communication, agglomeration and mapping pertaining to the partition model of medium granularity, are respectively analyzed in detail. In general, the importance of the partition model of medium granularity is as follows:(1) The model holds a connecting link between the preceding and the present. The profound comprehension on the parallel system software is achieved when the model is able to syncreticly absorb the preceding research and the present research on the parallel system software.(2) The model also holds a connecting link between the present and the future. The model has theoretical importance for the further research on the parallel system software. For example, according to the model partitioning practice, at present the nodus of developing parallel software is lying in the partitioning method of the model. Based on the above consideration, the partition of circulation is to be deeply researched.(2) Secondly, two new parallel algorithms are initiated to partition the common circulation mode in parallel by introducing the thoughts of the correlation set, the non-correlation set, the direct correlation chain and the group of direct correlation chains and depending on the theory of the great common divisor. One is defined as "the new partition algorithm of circulation based on the great common divisor", the other is defined as "the deep-seated partition algorithm of circulation" in this paper. Moreover,the programs for computer, corresponding to the two new algorithms respectively, have been designed. The computational results of the examples show that:(1) In terms of the new partition algorithm of circulation based on the great common divisor, there are many merits described as follows:The algorithm can eliminate the redundant computation of correlation points completely; the algorithm can obtain better load-balancing; the most impressive is that the algorithm can extend the application spectrum, where the theory of the great common divisor applies; the algorithm is simple in design, fast in computation, steady in performance and effective in practice; the algorithm is extensible and fit for the partitioning of multi-loop.(2) In terms of the new deep-seated partition algorithm of circulation, besides the merits above-mentioned, it is the most important that it is very effective to partition circulation, because it can partition the correlation set of circulation in parallel. Moreover, the new deep-seated partition algorithm of circulation can combine with the new partition algorithm of circulation based on the great common divisor to partition the common circulation mode in parallel.(3) Thirdly, the traditional implicit Runge-Kutta scheme with variable steps is partitioned in parallel based on the above-mentioned partition algorithms of circulation. The simulation results of the example show that the parallel implicit Runge-Kutta scheme with variable steps is rational in theory and effective in practice. The processing time of the parallel implicit Runge-Kutta scheme with variable steps will decrease from about 29% to 58%, compared to the total serial processing time of the traditional implicit Runge-Kutta scheme with variable steps.All the approaches here are signifi...