24小时热门版块排行榜

返回列表

sjnyongle

金虫 (小有名气)

应助: 0 (幼儿园)
金币: 402.5
帖子: 51
在线: 18.2小时
虫号: 1107240

[交流] 【求助】the usage of ++nodelist and ++local

各位大侠，请教一个问题，非常感谢回答：
Is NAMD_2.7_Linux-x86_64 only used in the Individual Linux Workstations, other than the infiniband network, even if I add ++nodelist in the command line?

This question initiates from Recently, I'm relating to using NAMD_2.7_Linux-x86_64 in the Linux cluster of infiniband Network.
When I sub my job, in the log file, there are just two lines:

Warning> Randomization of stack pointer is turned on in kernel, thread migration may not work! Run 'echo 0 > /proc/sys/kernel/randomize_va_space' as root to disable it, or try run with '+isomalloc_sync'.
Charm++: scheduler running in netpoll mode.

No more information write out in the log file. Then I change the last command line in job-sub-file to:

$NAMD/charmrun ++local +p${n_proc} $NAMD/namd2 2rh1.md > 2rh1_64cpu.log

It can run normally.

So, I'm confused about the usage of ++nodelist and ++local. Because my Linux cluster is infiniband network, then it may be right to use ++nodelist, but the ++nodelist can not work.
Furthermore, if I change to ++local, the speed of calculation is much slower than the NAMD_2.7_Linux-x86_64-ibverbs. Could anyone help me understand this?

My job-submitting file is:

>cat sub.sh
#!/bin/bash -x
#PBS -N 2rh1_nodelist_10
#PBS -l nodes=8:ppn=8
#PBS -j oe
#PBS -q parallel
#
#define variables
#
n_proc=$(cat $PBS_NODEFILE | wc -l)
echo $PBS_NODEFILE $n_proc
#
#running jobs
#
cd $PBS_O_WORKDIR
echo $PBS_O_WORKDIR
#
# Setup the MPI topology
#
export CONV_RSH=ssh
echo "group main" >node1
awk '{print "host ", $1}' $PBS_NODEFILE >>node1
export NAMD=/home/users/ymei/sjn/NAMD_2.7_Linux-x86_64
$NAMD/charmrun ++nodelist node1 +p${n_proc} $NAMD/namd2 2rh1.md > 2rh1_64cpu.log

回复此楼