Linux IA32下MPP970 revision 6763执行设定，Lam-mpi_7.0.3版

aptkevin

LSTC公司release的最新版本的MPP970为6763，新版本与5434a等老版本的执行方式有所变化。直接将下载的可执行文件替换旧版，然后用mpirun -np 2 n0 /mpp970_lam_revision i=keyfile.k memory=20m 来进行提交会出现找不到libf77*.a.0文件找不到的错误。新的6763版本还需要与操作平台相对应的share library文件，可以从
ftp://user@ftp.lstc.com/mpp-dyna/MPI_shared_libs
下载已经编译好的library文件，也可以自己下进行编译生成。自已编译生成库文件的方法可以参考如下：
(1) how to create MPI shared libraries and proper MPI environment:
    For example to create MPICH 1.2.6 on EM64t
    (a) using gcc and g77
       ./configure -cc=cc -fc=g77 -fflags=-fno-second-underscore \
                   -device=ch_p4 -comm=ch_p4 --enable-sharedlib  \
                   -opt=-O -prefix=/yourpath/mpich-1.2.6
    (b) using gcc and Intel Fortran compiler
       ./configure -cc=cc -fc=ifc -device=ch_p4 -comm=ch_p4 --enable-sharedlib \
                   -opt=-O -prefix=/yourpath/mpich-1.2.6
    After successfully configure the MPI, you need to compile the MPI source
    to generate all the utilities via
    gmake
    gmake install
    After completion, please change dir to ./lib/shared to check if all
    lib*.so have been created or not.
    Set up the LD_LIBRARY_PATH to include the shared libraries to it
    under C shell
    setenv LD_LIBRARY_PATH /yourpath/mpich-1.2.6/lib/shared:$LD_LIBRARY_PATH
    under B shell
    export LD_LIBRARY_PATH=/yourpath/mpich-1.2.6/lib/shared:$LD_LIBRARY_PATH
上面提到须将share library 的目录加到LD_LIBRARY_PATH环境变量中，但值得注意的是直接在current ENV里执行上面的命令，然后提交计算任务依然还是会出现错误。用$ldd -r mpp970命令也可以看到还是有几个参考的lib文件找不到。
官方release的安装指南里也没有提到更多细节，但从google得到了解释。参考如下文章
http://www.lam-mpi.org/MailArchives/lam/2005/04/10393.php
Michael,
Couple of things to keep in mind.
1. Is the library in your LD_LIBRARY_PATH.
2. Is the LD_LIBRARY_PATH setup in your .files or just in your current ENV?
You need to have the LD_LIBRARY_PATH point to the right location in your .files since the mpp version with use RSH_COMMAND to start a new thread and will not inherit your current ENV.. but create a new ENV based on your .files.
HTH,
Pete.
On Wednesday 20 April 2005 10:40, Evans, Michael wrote:
> I am trying to run the MPP version of LSDYNA (mpp970) on SUN but get the
> following error message: fatal: libsunperf.so.4: open failed: No such file
> or directory. There are 2 copies of the library on the server and at least
> one of them is in my path. Is there a certain version of this library I
> need or am I missing something else?
>
> Michael Evans
> UNIX Administrator
> ArvinMeritor
> Troy Computing Services
>
>
> _______________________________________________
> This list is archived at
http://www.lam-mpi.org/MailArchives/lam/

--
Peter Schmid
看来问题在于mpp970的启动利用rsh的方式，其环境变量不是继承当前的ENV，因此须将export LD_LIBRARY_PATH=/path/to/share/librarys设置在登陆脚本文件里。

本文来自ChinaUnix博客，如果查看原文请点：http://blog.chinaunix.net/u/3225/showart_76134.html