- 积分
- 19
- 注册时间
- 2007-1-25
- 仿真币
-
- 最后登录
- 1970-1-1
|
本帖最后由 monnet 于 2009-8-5 09:48 编辑
看到SimWe上有不少人对COMSOL的并行心存疑虑,重新翻到COMSOL News 2008里COMSOL技术支持Linus Andersson先生的一篇文章,希望对大家对大问题的计算有帮助。
The more processors you use for computing, the quicker your simulations run. By allowing for multicore processing of your COMSOL Multiphysics models, you can achieve a significant speedup in 3D meshing, assembly, and solving. Here are some tips to help you better leverage your multicore platform for multiphysics modeling.
To run COMSOL Multiphysics with multicore processor support on Linux or Sun platforms, use the command:
comsol -np n
Where n is the number of processor cores you want to use.
In Windows, COMSOL Multiphysics will use all available cores by default. You can use the same command if you want to reserve some cores for a separate COMSOL session or for other tasks.
In a multicore environment, COMSOL distributes meshing between cores subdomain by subdomain. To maximize performance, use more subdomains than the physics of your model would otherwise dictate. Say, for instance, that you have a quad-core processor, and you would like to speed up the meshing process in a model with a complex geometry but nominally only one subdomain. Try splitting the geometry into four subdomains, each with similar size and complexity, and then do your meshing.
In the solution process, the degree of parallelization mostly depends on which solver you use. Generally, the solver that benefits most from multicore processing is the PARDISO direct solver. All iterative solvers and smoothers except Incomplete LU are also parallelized. |
评分
-
1
查看全部评分
-
|