The Research Of Parallel Optimization Of The Multi-core Numerical Algorithm

With the increasing popularity of multi-core computers, large-scale scientific and engineering computing, the processing of transactions, and the needs of business computing, and application field of data mining expanding constantly, the research which is based on numerical solution algorithm of linear equations and the associated rule algorithm of the numerical value of large data of multi-core platforms is very important. As multi-core technology will increase the costs of the creation of multi-threaded, communication, and destruction, that how to find the parallel performance in the program, optimize the parallel program of multi-core and play the best performance of multi-core,have become the important issues in high-performance computing field.This paper presents a detailed derivation process of Cholesky decomposition algorithms, analyses tasks allocation strategy of the Cholesky decomposition parallel in accordance with block assigned for existing load imbalance, proposes the cross scheme of block allocation program and the scheme of shutter distribution, and adoptes advance transmission strategy for the three algorithms. This paper conducts experiments and analysis on the MPI cluster system. The complexity of parallel algorithm design based on MPI for Cholesky decomposition, the difficulties in the load balancing of process, problem of a big overhead of message passing and data movement, and the enlargement of parallelism overhead in using OpenMP parallelism and the problem of the thread-load imbalance based on multi-core, this paper pays attention to the advantage of multi-core computers, proposes multi-core parallel optimization algorithm based on a division of tasks of the Cholesky decomposition. The algorithm divides large circulation problems into small independent tasks each other and uses the recursive technology, task stealing technology and dynamic load balancing algorithm so that these tasks can be done in parallel.The paper analyses the Apriori algorithm of the association rule mining and shortcomings of its parallel algorithm existence, proposes an improved an optimization algorithm of multi-core parallel of the association rule mining. This algorithm transformes the compression matrix of Apriori algorithm, and uses OpenMP and TBB technology on the multi-core platform to conduct serial program into cycle parallelization design and parallelization design of task allocation, realizes the parallel association rule mining ,improves the efficiency of association rules mining and utilization of multi-core CPU.This paper introduces the parallel suit of Intel Parallel Studio in the design algorithm program, checks the memory and thread and optimizes performance based on multi-core proposing parallel algorithms, eliminates the problem which the use of multi-core platform parallel resulted in data leakage, data competition, deadlock and low efficiency. Experimental results show that the proposed algorithm based on multi-core has high efficiency and good parallel speedup.