Research On The Construction And Analysis Of Brain Network Based On Heterogeneous Parallel System

The human brain is one of the most complex networks in nature, and complex network theory provides a new direction for the research of the human brain. Brain network attribute calculation is an important way to study the brain network. So the research time of network is the most important factor to affect the progress of the research. This is a important indice to provide technical feasibility for the brain study. The development of the brain network makes the network granularity increasingly fine, the scale of network is becoming larger,and the computation presents the geometric progression growth. The time of brain network research is too long to meet the actual needs of the research,greatly hindered the process of brain network. With the development of parallel computing technology, it has been a very effective means for the data intensive computing tasks. Therefore, this paper is mainly to use parallel computing technologies to achieve the fast computation of the brain network.The focus of this paper is a parallel implementation of calculation of the resting state brain network. With using the relevant theory of parallel computing,the parallel strategies are proposed for the brain network research in differentparallel computing platforms. This paper brings some new ideas:(1) Based on the heterogeneous environment of CPU-GPU, the paper proposes a new method to realize the construction of the brain network. On the basis of this, a parallel strategy is proposed for the research of the brain network,which is to use the CUDA architecture and the CUBLAS library to speed up the calculation module. The performance of the parallel strategy is tested on a parallel platform which supports GPU general computing. The optimal parallel effect is to achieve a 3 times speedup.(2) In parallel system based on multi-CPU environment, a parallel strategy for brain network computation is proposed. That is, the system uses the multi-CPU properties to compute multiple brain networks at the same time. In the parallel platform supporting multi CPU computing, the architecture based on SPMD mechanism and cyclic packing method is proposed, and the parallel performance of this kind of parallel strategy is tested, and the maximum speedup is more than 6 times.(3) Based on CPU-GPU heterogeneous parallel system, this paper presents the combination of the above two kinds of strategies. It is also the combination of the heterogeneous environment of multi-CPU and GPU to implement the seamless operation. In view of this kind of combination strategy, implement this parallel method and test its performance. The best parallel effect is obtained by nearly 4 times speedup.