Research Of Multicore Runtime Scheduling System For Parallel Power Calculation

With the development of the power grid size, the power system becomes more complex, it becomes difficult to estimate the state of the system. The real-time simulation and fast computing for power calculations put forward higher requirements, which require scientific parallel computing to solve. So the parallel computing power calculation becomes very necessary. Traditional methods can achieve a simple parallel computing in power system and improve certain performance. But for the irregular programs, load imbalance and a lot of parallel overhead are easily caused by the traditional methods, which will be difficult to take the advantages of parallel computing.The paper studies the parallelization of a power calculation algorithm called Prony algorithm and implements it on multi-cores system scheduled by Adaptive Task Granularity Control System(ATGCS). In order to achieve high efficiency in parallel computing and improve the rate of response, the parallel Prony algorithm based on ATGCS uses adaptive task granularity control to create parallel tasks. On the one hand, it reduces the cost of parallel process. On the other hand, by using the Work-Stealing strategy to steal tasks between different task queues, it reduces the overhead of tasks' creation and execution.Multicore Runtime Scheduling System for Parallel Power Calculation achieves a first complete parallelization of Prony algorithm. We also make a comparison of Parallel Prony algorithm based on OpenMP and Parallel Prony algorithm based on ATGCS. Experiment results show that parallel Prony algorithm based on ATGCS can achieve better speedup compared with Parallel Prony algorithm based on OpenMP. ATGCS is more suitable for parallel of Prony algorithm, and can improve the performance by up to 20% compared with Parallel Prony algorithm based on OpenMP.