我们可以在单个系统中运行 MPI 程序还是必须仅在集群中运行它?
我可以使用 PelicanHPC 访问大学的集群网络,其中运行各种 MPI 程序,但在家里我想练习编写/使用其他 MPI 程序。有没有办法可以在我自己的系统上运行 MPI 程序?
(我在 Ubuntu Jaunty 上工作)
I have access to a clustered network at my college using PelicanHPC where In run various MPI programs, but at home I want to practice writing/using other MPI programs. Is there a way that I can run MPI programs on my own system?
(I work on Ubuntu Jaunty)
如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱,如果其他用户或者作者回复了您的评论,将不能在第一时间通知您!
发布评论
评论(2)
因此,根据 http://idea.uab.es/mcreel/ParallelKnoppix/, PelicanHPC“已安装 MPI 的 LAM-MPI 和 OpenMPI 实现。”
我不了解 LAM-MPI,但我知道 OpenMPI 会自动平衡多个处理器之间的线程,只要您不要求比处理器更多的线程。这意味着,使用双核计算机,您可以随心所欲地“mpirun -n 2”。但是,如果您想以真正的并行性“mpirun -n 8”,则需要 4 个双核盒子。
So according to http://idea.uab.es/mcreel/ParallelKnoppix/, on PelicanHPC "The LAM-MPI and OpenMPI implementations of MPI are installed."
I don't know about LAM-MPI, but I know OpenMPI will automatically balance threads across multiple processors, as long as you don't ask for more threads than processors. This means that with a dual-core computer, you can "mpirun -n 2" to your heart's content. However, if you want to "mpirun -n 8" with true parallelism, you'd need 4 dual-core boxes.
这将取决于您使用的 MPI 运行时(您将需要运行时 - 类似于 mpich )。我想在任何情况下您都可以在多个进程中运行该程序,但是如果您运行更多进程,那么您拥有的处理器核心就会减少并行度。
This will depend on the MPI runtime you use (you will need the runtime - something like mpich). I guess in any case you can run the program in several processes, but if you run more processes then you have processor cores you will have less parallelism.