Cluster Installation

XDS can be run in cluster mode using any command line job scheduling software such as Grid Engine, Condor, Torque/PBS, LSF, SLURM etc. We implemented Grid Engine. It is a distributed resource management system which monitors the CPU and memory usage of the available computing resources and schedules the job to the least used computer. Grid Engine was chosen due to its high scalability, cost effectiveness, ease of maintenance and high throughput. Grid Engine was developed by Sun Microsystems (Sun Grid Engine, SGE) and later acquired by Oracle and subsequently acquired by UNIVA. The latest versions became closed source, but the older ones are open source supplied with many Linux distributions including Redhat/CentOS 6.x. There is also open source Open Grid Scheduler [[1]], Son of Gridengine [[2]]

XDS Cluster setup

In order to setup XDS in cluster mode, forkcolspot and forkintegrate scripts need to be changed to access the gridengine environment and send jobs to different machines. Example scripts are below, need to be changed according to the environment. Observe there is qsub command which submits forkcolspot_job/forkintegrate_job to grid engine.

#forkcolspot

ntask=$1  #total number of jobs
maxcpu=$2 #maximum number of processors used by each job
	   #maxcpu=1: use 'mcolspot' (single processor)
	   #maxcpu>1: use 'mcolspot_par' (openmp version)

pids=""                    #list of background process ID's
itask=1
echo "MAX CPU $maxcpu $image1"

#check for gridengine submit host
submitnodes=`qconf -sh 2> /dev/null`
thishost=`hostname`
isgrid=0
for node in $submitnodes ; do
	if [ "$node" == "$thishost" ]
	then
		isgrid=1
		echo "Grid Engine environment detected"
	fi
done

while test $itask -le $ntask
do
   if [ $maxcpu -gt 1 ]
#     then echo "$itask" | mcolspot_par &
#     else echo "$itask" | mcolspot     &
      then 
      	if [ $isgrid -eq 1 ]
	then
        #submit job to grid engine
	qsub -sync y -V -l h_rt=0:20:00 -cwd \
	  forkcolspot_job \
	  $itask  &
	 
      	#else echo "$itask" | qrsh -V -cwd "mcolspot"     &
	else echo "$itask" | mcolspot_par &
	fi 
   else echo "$itask" | mcolspot     &	
   fi
   pids="$pids $!"  #append id of the background process just started

   itask=`expr $itask + 1`
done
trap "kill -15 $pids" 2 15  # 2:Control-C; 15:kill
wait  #wait for all background processes issued by this shell
rm -f mcolspot.tmp  #this temporary file was generated by xds
rm -rf fork*job*

#forkcolspot_job

#!/bin/csh

echo $1
set itask=$1
set host=`uname -a | awk '{print $2}'`
echo $itask $host >> jobs.log
echo $itask | mcolspot_par


#forkintegate

fframe=$1 #id number of the first image
ni=$2     #number of images in the data set
ntask=$3  #total number of jobs
niba0=$4  #minimum number of images in a batch
maxcpu=$5 #maximum number of processors used by each job
          #maxcpu=1: use 'mintegrate' (single processor)
          #maxcpu>1: use 'mintegrate_par' (openmp version)

minitask=$(($ni / $ntask)) #minimum number of images in a job
mtask=$(($ni % $ntask))    #number of jobs with minitask+1 images
pids=""                    #list of background process ID's
nba=0
litask=0
itask=1

#Sudhir check for gridengine submit host
submitnodes=`qconf -sh 2> /dev/null`
thishost=`hostname`
isgrid=0
for node in $submitnodes ; do
	if [ "$node" == "$thishost" ]
	then
		isgrid=1
		echo "Grid Engine environment detected"
	fi
done

while test $itask -le $ntask
do
   if [ $itask -gt $mtask ]
      then nitask=$minitask
      else nitask=$(($minitask + 1))
   fi
   fitask=`expr $litask + 1`
   litask=`expr $litask + $nitask`
   if [ $nitask -lt $niba0 ]
      then n=$nitask
      else n=$niba0
   fi
   if [ $n -lt 1 ]
      then n=1
   fi
   nbatask=$(($nitask / $n))
   nba=`expr $nba + $nbatask`
   image1=$(($fframe + $fitask - 1)) #id number of the first image

   if [ $maxcpu -gt 1 ]
      then 
      if [ $isgrid -eq 1 ]
	then
        #submit job to grid engine
      	qsub -sync y -V -l h_rt=0:20:00 -cwd \
	  forkintegrate_job \
	  $image1 $nitask $itask $nbatask &
      #else echo "$image1 $nitask $itask $nbatask" | qrsh -V -cwd "mintegrate"     &
      else echo "$image1 $nitask $itask $nbatask" | mintegrate_par  &
      fi
      else echo "$image1 $nitask $itask $nbatask" | mintegrate  &
   fi
   pids="$pids $!"  #append id of the background process just started

   itask=`expr $itask + 1`
done
trap "kill -15 $pids" 2 15  # 2:Control-C; 15:kill
wait  #wait for all background processes issued by this shell
rm -f mintegrate.tmp  #this temporary file was generated by mintegrate
rm -rf fork*job*
#forkintegrate_job

#!/bin/csh

set image1=$1
set nitask=$2
set itask=$3
set nbatask=$4

set host=`uname -a | awk '{print $2}'`
echo $image1 $nitask $itask $nbatask $host >> jobs.log
echo $image1 $nitask $itask $nbatask | mintegrate_par

Grid Engine Installation

Grid Engine consists of a master node daemon named sgemaster which schedules jobs to execution nodes. On each execution node a daemon named sge_execd runs a job and sends a completion signal back to sgemaster. Jobs are submitted to sgemaster using command such as qsub or using DRMAA C, JAVA or IDL bindings from any applications which want to run XDS.

 


Redhas/CentOS Linux distribution comes with rpms for installing Grid Engine. One need to have administrative privileges to install. Install gridengine rpms on all the nodes using following command, Default shell for Grid Engine is /bin/csh. It is assumed that all the workstations involved access the storage (using NFS or other cluster file systems) where the data is stored and authentication is done through protocols like LDAP.

root@ws1:/home 1> yum install gridengine gridengine-qmaster gridengine-execd  gridengine-qmon

root@ws1:/home 2> rpm -qa | grep gridengine

gridengine-qmaster-6.2u5-10.el6.4.x86_64
gridengine-qmon-6.2u5-10.el6.4.x86_64
gridengine-execd-6.2u5-10.el6.4.x86_64
gridengine-6.2u5-10.el6.4.x86_64

By default gridengine installation directory /usr/share/gridengine, contents shown below.

root@ws1:/home 3> cd /usr/share/gridengine

root@ws1:/home 4> ls
bin   default  hadoop    install_execd    lib  my_configuration.conf  qmon  utilbin
ckpt  doc      inst_sge  install_qmaster  mpi  pvm                    util

Lets say ws1 is sgemaster node, it will installed using install_qmaster

Installing sgemaster

root@ws1:/usr/share/gridengine 5>./install_qmaster

Most of the answers are yes/no or typing enter. Following things need to be decided before installation

  • Admin user is root
  • Following important environment variables are written to /usr/share/gridengine/default/common/settings.csh which should be in the $PATH.
    • $SGE_ROOT=/usr/share/gridengine
    • $SGE_QMASTER_PORT=6444
    • $SGE_EXECD_PORT=6445
    • $SGE_CELL=default
  • JMX MBean server not used
  • Spooling method used is classic
  • There is an option to give administrative email which is very useful, when ever there is any problem gridengine will send error messages to email.
  • Ready with a file contains admin and submit hosts or you can manually enter all the hosts separated by space, use full DNS names of hosts.
  • In this installation shadow host is not used.
  • After the shadow host step make sure allhosts group and all.q are created otherwise installation sge_execd will have problems.
  • Scheduler Tuning selected as 'Max', it has disadvantage, gridengine immediately schedules with out assuming the load, this will cause successive job submissions will go to same host until all the slots are filled for that machine. Selecting 'Normal' will assume the load but there is overhead of few sec. extra time for job scheduling.


After finishing the installation the configuration files are automatically written to the directory /usr/share/gridengine/default since the cell name selected is 'default'. This directory can be choosen as a shared directory over NFS. Otherwise copy this directory to every host used in the cluster.

Installing sge_execd

On execution node install execution daemon using following command

root@ws2:/usr/share/gridengine 5>./install_execd

the input is almost typing return if you already copied the 'default' directory to this node.


Submit nodes

Install Grid engine rpms also on all the submit nodes which use qsub, and copy the default directory /usr/share/gridengine/default. Remember all the execution nodes in the cluster also act as submit nodes in case of XDS.

Use command qconf to see which are submit hosts which are not and you can add them manually.


Restarting Grid Engine

When grid engine installed first time /etc/init.d/sgemaster and /etc/init.d/sge_execd services are automatically installed. If you want to restart sgemaster make sure all the sge_execd deamons are stoped. You can do this by following commands

service sge_execd stop
service sgemaster stop

for starting

service sge_execd start
service sgemaster start

When ever work stations need to be restarted make sure sgemaster work station started first. To keep the services restarted automatically during the startup make sure chkconfig is on.

chkconfig sgemaster on
chkconfig sge_execd on

Son of Gridengine

rpms available in this link

http://arc.liv.ac.uk/downloads/SGE/releases/8.1.8/

by defualt these rpms install in single directory /opt/sge instead of scattering files (by default) to /usr/bin, /usr/share/gridengine, /usr/spool/gridengine

Default shell for Son of Gridengine is /bin/sh which is /bin/bash