单机并行计算，并行的和越多，运算速度越慢，为什么呢？

单机并行计算，并行的和越多，运算速度越慢，为什么呢？

大家好，我做单机多核并行计算后，速度反而没有一个跑得快了，而且并行的核越多，速度越慢。
    软件是lammps，并行软件是mpich。跑得原子数是24000个。我看了下log文件，多核相对于单核来说，kspace time和comm time占的时间比例更多了。
    不知道有没有说清楚，期待大侠回复。。。
哈哈，该问题解决啦！主要是师兄的帮忙，不然，肯定是搞不定的。在这里，和大家分享一下问题的根源
主要是因为Linux的内核有两种，一是.elsmp，一是el，前者是symmetric multiprocessing对称对处理器，后者是uniprocessor单处理器。而我启动Linux时，选择的是后者，所以并行的核越多，速度越慢。因此，要并行的话，应该选择elsmp才对。
其实，问题很简单，呵呵，高手不要笑我啊。。。。。:)

solve fluent problem

ok, so today learn how to fix the error in fluent due to parallel processing,

 http://www.cfd-online.com/Forums/fluent/85262-error-fluent-parallel-processing.html

I don't mean to bring up a thread that hasn't had any new posts in over a month, but I found this by googling (first hit) and assume others might as well.

I'm using Fluent from Ansys V13 and had a similar issue where the program would hang while trying to spawn nodes. I had installed Intel MPI (not HP, but it's possibly similar) and to fix the problem, I ran wmpiregister. I'd assume if you installed HP MPI it would be similar. You have to register your username for windows and give it a password (which I believe must match between windows and whatever mpi you're using). 

Then, when starting Fluent, under the more options section/parallel settings tab, under MPI types I denoted that I am using Intel MPI instead of default.

Works fine now and no more hanging. Hope that helps anyone who finds this.