Research Of Multi-core Program Optimization Based On Task Parallel Strategies

With the popularity of the multi-core processor.How to effectively use the multi-core processor resources, become a research hotspot of parallel computation field. Now many applications are developed in mononuclear environment.In multi-core environments execution with single thread, led to many waste of multi-core resources. In order to give full play to multi-core potential and ascending application performance.Analysis of parallel technology under the premise of multi-core , research of multi-core program optimization based on task parallel strategies . The advantage of task parallel is lightweight components, Start and end task of spending small than to create and distribution of the threads.Through studying the Intel thread building blocks (TBB), This paper puts forward the parallel optimization methods based on task scheduler,.This method is through the task scheduler create a thread pool management tasks,using task stealing technology load balancing, The task automatically mapping to multiple logical thread parallel execution. So as to improve the utilization efficiency of multicore resources and the performance of the program. For multi-core clusters environment made a multilayer task to parallelize method, Provide upper between the nodes MPI parallel process task, The lower TBB provide multi-core node internal task parallel. The advantage of both, achieve progress and processing nodes,threads and core the two levels of parallel processing. In order to validate the efficiency of two methods mentioned above, Classical algorithm with N queen as an example, In a Shared storage model TBB and mixed parallel programming model environment TBB +MPI runing the iterative parallelization test. Experimental results show that sharing storage model parallelization speedup ratio and CPU utilization can achieve a considerable numerical, reflected the task scheduler parallelization superiority.,and obviously improve the cluster parallel performance.Finally based on fuzzy c-means algorithm as an example, using the OpenMP work division coding, realize the serial program hot spots task parallelism. Intel studio parallelize of high performance tools to parallelize results were verified, Comparison the results of execution time and concurrent, its performance has improved a lot. Confirm the task of parallel strategies stality and realibility characteristic,it is a good practicability for mang multi-core application performance optimization .