应《网络安全法》要求，自2017年10月1日起，未进行实名认证将不得使用互联网跟帖服务。为保障您的帐号能够正常使用，请尽快对帐号进行手机号验证，感谢您的理解与支持！

24小时热门版块排行榜    

小木虫论坛-学术科研互动平台 » 计算模拟区 » 量子化学 » Linux应用 » mpirun并行运行问题
51/1返回列表
查看: 1861  |  回复: 3
只看楼主 @他人 存档 新回复提醒 (忽略) 收藏    在APP中查看
当前只显示满足指定条件的回帖，点击这里查看本话题的所有回帖

刘志凌

木虫 (小有名气)

[求助] mpirun并行运行问题

我在集群上运行一个软件,本意要求它一次同时运行两个任务,但是每次它都只是运行一个任务,软件的作者解释是如下,不知有没有大侠能否指点一下迷津.先行谢谢了.
It appears that both mpi processes launched, but for some reason they launched independently of each other.  This means that neither process was aware of the other process, so both processes were doing the same things.  Both processes opened Test.inp and then tried to create this directory:

/home/Test/Gaussian/bestSavedStructures

One process succeded, and the other failed.  The one that failed printed "Couldn't create log files directory." in the .e file and immediately stopped.  The other kept going and made it to iteration 10, which was the predetermined stopping point.  There didn't seem to be any problems with the disk which is very good.  If you can figure out how to get the processes to see each other, you'll have a working program.

Upon launch, each process prints it's rank (or index) and the total number of processes.  This is what was printed during this run:
This is MPI process with rank 0.  Total MPI processes: 1
This is MPI process with rank 0.  Total MPI processes: 1

This is what it should have printed:
This is MPI process with rank 0.  Total MPI processes: 2
This is MPI process with rank 1.  Total MPI processes: 2

Please replace '-np 2' in pso.pbs with another option.  You might try '-hostfile nodes.txt' or '-cpus-per-rank 1'.  Or look at other options listed at http://www.open-mpi.org/doc/v1.4/man1/mpirun.1.php .  Hope that helps!
回复此楼

» 猜你喜欢

» 本主题相关价值贴推荐，对您同样有帮助:
好人一生平安
1楼 2012-10-29 20:31:29
已阅   关注TA 给TA发消息 送TA红花 TA的回帖

virtualzx

木虫 (著名写手)
引用回帖:
2楼: Originally posted by 刘志凌 at 2012-10-29 07:32:28
作者的指示的那几个选项我都试过了,结果还是一次运行一个任务

请给出你的命令行
赞一下 回复此楼
3楼2012-10-30 03:53:15
已阅   关注TA 给TA发消息 送TA红花 TA的回帖
查看全部 4 个回答

刘志凌

木虫 (小有名气)
作者的指示的那几个选项我都试过了,结果还是一次运行一个任务
赞一下 回复此楼
好人一生平安
2楼2012-10-29 20:32:28
已阅   关注TA 给TA发消息 送TA红花 TA的回帖

刘志凌

木虫 (小有名气)
引用回帖:
3楼: Originally posted by virtualzx at 2012-10-30 03:53:15
请给出你的命令行...

#!/bin/bash
#PBS -N TestJob
#PBS -l nodes=2:ppn=8,walltime=0:30:00
#PBS -l mem=4GB
export LD_LIBRARY_PATH=/home/fan_group/openmpi/lib:$LD_LIBRARY_PATH
echo PBS_O_WORKDIR=$PBS_O_WORKDIR
echo PBS_ENVIRONMENT=$PBS_ENVIRONMENT
echo PBS_NODEFILE=$PBS_NODEFILE
echo cat $PBS_NODEFILE
cat $PBS_NODEFILE
echo $PBS_NODEFILE
echo HOSTNAME=`/bin/hostname`

# Create the nodes.txt file
# The next line writes one node per line and allows duplicate nodes
# Use this if you want one Gaussian job per processor.
# cp $PBS_NODEFILE /home/fan_group/lzhl/pso/Test/Gaussian/nodes.txt


# The next line writes one node per line and does NOT allow duplicate nodes.
# Use this if you want one Gaussian job per node with each job runing in parallel on multiple processors.
cat $PBS_NODEFILE | uniq > /home/fan_group/lzhl/pso/Test/Gaussian/nodes.txt

cd /home/fan_group/pso1.5.2
/opt/openmpi/bin/mpirun -np 2 ./pso -walltime 0:30:00 /home/fan_group/lzhl/pso/Test/Gaussian/Test.inp
赞一下 回复此楼
好人一生平安
4楼2012-11-05 10:30:44
已阅   关注TA 给TA发消息 送TA红花 TA的回帖
查看全部 4 个回答
信息提示
关闭
请填处理意见
关闭