Changeset 635 for palm/trunk/SCRIPTS/subjob

Timestamp:

Dec 14, 2010 1:33:02 PM (14 years ago)

Author:

raasch

Message:

mrun/mbuild/subjob adjustments for new Tsubama system at Tokyo Institute of Techology (lctit)

File:

• palm/trunk/SCRIPTS/subjob (modified) (13 diffs)

palm/trunk/SCRIPTS/subjob

@@ -124 +124 @@
      #                    SGI-ICE/lcsgi
      #                    adjustments for Kyushu Univ. (lcrte, ibmku)
+     # 14/12/10 - Siggi - adjustments for new Tsubame system at Tokyo
+     #                    institute of technology (lctit)
 
 
@@ -236 +238 @@
      (r1*)                   local_addres=130.75.4.102;   local_host=lcsgih;;
      (r2*)                   local_addres=130.73.232.102; local_host=lcsgib;;
-     (tgg*)                  local_addres=172.17.75.161;  local_host=lctit;;
+     (t2a*)                  local_addres=10.1.6.165;     local_host=lctit;;
      (vorias)                local_addres=172.20.25.43;   local_host=lcmuk;;
      (*.cc.kyushu-u.ac.jp)   local_addres=133.5.4.129;    local_host=ibmku;;
@@ -367 +369 @@
         (lcsgih)  queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
         (lck)     remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
-        (lctit)   queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
+        (lctit)   queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
         (lcxt4)   remote_addres=129.177.20.113; submcom=/opt/torque/2.4.9-snap.201005191035/bin/qsub;;
         (lcxt5m)  remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
@@ -415 +417 @@
                  esac;;
         (lctit)  case  $ndq  in
-                     (novice|sla1|sla2|sla3|RAM64GB|RAM128GB)    error=false;;
+                     (G|L128|L256|L512H|S|S96|V)             error=false;;
                      (*)                                     error=true;;
                  esac;;
@@ -513 +515 @@
        (( Memory = memory / 1000 ))
     fi
+ elif [[ $remote_host = lctit ]]
+ then
+    (( Memory = memory * tasks_per_node / 1000 ))
  fi
 
@@ -1031 +1036 @@
  then
     cat > $job_to_send << %%END%%
-#!/bin/bash
-#$ -S /bin/bash
-cd $job_catalog
-export PATH=$PALM_BIN:\$PATH
-echo \$PATH
+#!/bin/ksh
 
 %%END%%
 
        # OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
-  #  submcom="$submcom  -N $job_name  -sgeout $remote_dayfile  -q default"
-    submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile  -q $node_usage "
+    submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -l place=scatter -o $remote_dayfile -j oe -q $queue "
 
  else
@@ -1066 +1066 @@
     echo "trap '"                               >>  $job_to_send
     echo "set +vx"                              >>  $job_to_send
-    if [[ $(echo $remote_host | cut -c1-3) = ibm  ||  $(echo $remote_host | cut -c1-5) = lcsgi  ||  $(echo $remote_host | cut -c1-3) = nec  ||  $remote_host = lctit ]]
+    if [[ $(echo $remote_host | cut -c1-3) = ibm  ||  $(echo $remote_host | cut -c1-5) = lcsgi  ||  $(echo $remote_host | cut -c1-3) = nec ]]
     then
        if [[ $remote_host = ibmh ]]
@@ -1083 +1083 @@
        then
           return_queue=serialq
-       elif [[ $remote_host = lctit ]]
-       then
-          return_queue=sla3
        elif [[ $remote_host = necriam ]]
        then
@@ -1169 +1166 @@
           echo "[[ \"\$for_subjob_to_do\" != \"\" ]]  &&  eval \$for_subjob_to_do"  >>  $job_to_send
           echo "%%END%%"                           >>  $job_to_send
-
-       elif [[ $remote_host = lctit ]]
-       then
-          echo "cat > scpjob.$kennung << %%END%%"          >>  $job_to_send
-          echo "set -x"                                    >>  $job_to_send
-          echo "PATH=\$PATH:$job_catalog/../pub"           >>  $job_to_send
-          echo "cd $job_catalog"                           >>  $job_to_send
-          echo "batch_scp  -d  -w 10  -u $local_user $local_addres  $remote_dayfile  \"$job_catalog\"  $local_dayfile"  >>  $job_to_send
-          echo "[[ \"\$for_subjob_to_do\" != \"\" ]]  &&  eval \$for_subjob_to_do"  >>  $job_to_send
-          echo "%%END%%"                                   >>  $job_to_send
 
        else
@@ -1213 +1200 @@
           echo "chmod  u+x  scpjob.$kennung"    >>  $job_to_send
           echo "msub  scpjob.$kennung"          >>  $job_to_send
-       elif [[ $remote_host = lctit ]]
-       then
-          echo "chmod  u+x  scpjob.$kennung"    >>  $job_to_send
-          echo "rm -rf ${job_catalog}/last_job_transfer_protocol"  >>  $job_to_send
-          echo "n1ge  -N jobtransfer  -q $return_queue  -sgeout ${job_catalog}/last_job_transfer_protocol  scpjob.$kennung"  >>  $job_to_send
-          echo "sleep 10"                       >>  $job_to_send
        elif [[ $remote_host = t3eb  ||  $remote_host = t3eh  ||  $remote_host = t3ej2  ||  $remote_host = t3ej5 ]]
        then
@@ -1255 +1236 @@
     echo "exit"      >>  $job_to_send
  fi
+
+    # remove job file
  if [[ $remote_host = lctit  ||  $remote_host = ibmku ]]
  then
@@ -1322 +1305 @@
     then
        [[ $verify = true ]]  &&  printf "\n >>> submitting job using \"qsub\"...\n"
-       if [[ $remote_host != lctit ]]
-       then
-          if [[ $(echo $remote_host | cut -c1-5) = lcsgi  &&  $prio = true ]]
-          then
-             printf "\n >>> submit with HLRN qos-feature hiprio...\n"
-             ssh  $remote_addres  -l $remote_user  "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
-          elif [[ $remote_host = ibmku ]]
-          then
-             ssh  $remote_addres  -l $remote_user  "cd $job_catalog; $submcom $job_on_remhost"
-          else
-             ssh  $remote_addres  -l $remote_user  "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
-          fi
+
+       if [[ $(echo $remote_host | cut -c1-5) = lcsgi  &&  $prio = true ]]
+       then
+          printf "\n >>> submit with HLRN qos-feature hiprio...\n"
+          ssh  $remote_addres  -l $remote_user  "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
+       elif [[ $remote_host = ibmku ]]
+       then
+          ssh  $remote_addres  -l $remote_user  "cd $job_catalog; $submcom $job_on_remhost"
        else
-             # TIT ERLAUBT NUR DIE AUSFUEHRUNG GANZ BESTIMMTER KOMMANDOS
-             # MIT SSH, DESHALB AUFRUF PER PIPE
-             # UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
-          print  "cd $job_catalog; chmod u+x $job_on_remhost"  |  ssh  $remote_addres  -l $remote_user  > /dev/null  2>&1 
-          echo "first try" > send_protocol
-          while [[ $(cat send_protocol | grep -c "Forwarding to N1GE") = 0 ]]
-          do
-             if [[ $(cat send_protocol | grep -c "first try") = 1 ]]
-             then
-                printf "\n trying to submit job to TIT cluster..."
-             else
-                printf "\n +++ failed ... trying to submit again ..."
-             fi
-             print  "cd $job_catalog; $submcom $job_on_remhost"  |  ssh  $remote_addres  -l $remote_user  >  send_protocol  2>&1 
-             cat send_protocol
-          done
-          sleep 10
-          print  "cd $job_catalog;  rm $job_on_remhost"  |  ssh  $remote_addres  -l $remote_user  > /dev/null  2>&1 
-          echo "$submcom  $job_on_remhost"
-          rm  send_protocol
+          ssh  $remote_addres  -l $remote_user  "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
        fi
 
@@ -1365 +1325 @@
        elif [[  $local_host = lcfimm  ||  $local_host = lctit  ||  $localhost = lcxt4 || $localhost = lck ]]
        then
+          chmod  u+x  $job_on_remhost
+          echo "$submcom  $job_on_remhost"
           eval  $submcom  $job_on_remhost
-          echo "$submcom  $job_on_remhost"
-          chmod  u+x  $job_on_remhost
-
        elif [[ $local_host = nech ]]
        then