Parallelization And I/O Optimization For A 3-D Plasma Simulation Code

LARED-P, a three-dimensional laser plasma electromagnitic simulation code, largely simulates the interaction of intense laser and plasma. It uses partical-in-cell simulation method and needs immense computing resource, one CPU can do nothing about the simulation. In this paper, LARED-P is parallelized based on MPI paradigm for distributed memory parallel computing environment. The large scale (10 particles and 10 grid points) numerical simulation of 3-D laser-plasma is implemented successfully. High speedup ratio and good scalability are achieved, the parallel efficiency of 512 processors is 75%. On the other hand, To solve some problems about output of the results of LARED-P, the storage with real-time compression and high performance parallel I/O are implemented for the performance optimization of the post processing which decreases the cost of memory and CPU time. Key-Words: parallelization, numerical simulation, post processing, data management, I/O performance, plasma 1 Physical issue and mathematical principle The super-strong self-generation magnetic field generated by the interaction of the laser and plasma has great effect on many physical processes, such as inertial confinement nuclear fusion, plasma wave acceleration of particles, energy absorption, heat conduction, dynamic behavior of ions or electrons, plasma expansion and so on. The plasma simulation method simulates the characteristics of plasma by tracking the motion of a large number of charged particles under the action of self-consistent field and externally applied electromagnetic field. The limited particle size model and particle cloud grid method are put forward to reduce the amount of particles without losing the plasma collective behavior . The particle cloud grid method requires solving the particle relativity Newton-Lorentz equation (1), particle cloud equation (2) and the plasma electromagnetic Maxwell equation group (3). Where the particle motion equation is based on the Lagrange coordinates, the field equation is based on the Euler coordinates, and the particle cloud equation takes charge of the interaction of variables in two different coordinates.