The effective parallelization of processing exploiting the MPI library for the numerically exact quantum transfer matrix (QTM) and exact diagonalization (ED) deterministic simulations of chromium-based rings is propos...
详细信息
ISBN:
(数字)9783642551956
ISBN:
(纸本)9783642551956
The effective parallelization of processing exploiting the MPI library for the numerically exact quantum transfer matrix (QTM) and exact diagonalization (ED) deterministic simulations of chromium-based rings is proposed. In the QTM technique we have exploited parallelization of summation in the partition function. The efflciency of the QTM calculations is above 80% up to about 1000 processes. With our test programs we calculated low temperature torque, specific heat and entropy for the chromium ring Cr-8 exploiting realistic Hamiltonian with singleion anisotropy and the alternation of the nearest neighbor exchange couplings. Our parallelized ED technique makes use of the self-scheduling scheme and the longest processing time algorithm to distribute and diagonalize separate blocks of a Hamiltonian matrix by slave processes. Its parallel processing scales very well, with efflciency above 90% up to about 10 processes only. This scheme is improved by processing more input data sets in one job which leads to very good scalability up to arbitrary number of processes. The scaling is improved for both techniques when larger systems are considered.
A parallelization scheme, which drives processing in simulations of the Monte Carlo type, suitable in highly heterogeneous computer system of a general purpose, is proposed. The message passing is applied and the MPI ...
详细信息
ISBN:
(纸本)9783540681052
A parallelization scheme, which drives processing in simulations of the Monte Carlo type, suitable in highly heterogeneous computer system of a general purpose, is proposed. The message passing is applied and the MPI library is exploited. For testing, the 2D Ising model in a magnetic field is taken. The dependence of speedup on the number of parallel processes is studied, showing that the scheme works well in different parallel computer systems. The condition for the best speedup in these simulations is explained. The possibility of parallel use of any available computing power from the surrounding is also indicated.
暂无评论