GAMESS semaphores

mkdocs-project-dir/docs/Software_Guides/Other_Software.md

@@ -359,7 +359,56 @@ module load gamess/5Dec2014_R1/intel-2015-update2
 # By default, the module sets it to ~/Scratch/gamess/randomLabel
 export GAMESS_USERSCR=$TMPDIR
 
+# removes any of the user’s semaphore arrays that were left from previous
+# jobs on this node. Use if you are using a whole number of nodes.
+ipcrm -a sem
+
 rungms exam01.inp 00 $NSLOTS $(ppn)
+
+# removes all of the user’s semaphore arrays. 
+# Use if you are using a whole number of nodes.
+ipcrm -a sem
+```
+
+#### Semaphores
+
+GAMESS uses semaphore arrays and leaves these on the nodes where it has been running. This can 
+cause problems for your next jobs or for jobs from other users. 
+
+You will get errors like this if there are not enough semaphores available:
+
+```
+DDI Process 12: semget return an error.
+semget errno=ENOSPC -- check system limit for sysv semaphores.
+```
+
+These lines in the script above will clean up the semaphores belonging to your user on the
+node that you are on:
+
+```
+# removes all of the user’s semaphore arrays.
+# Use if you are using a whole number of nodes.
+ipcrm -a sem
+```
+
+Putting this before and after your GAMESS run ought to make sure that no semaphores from you are
+left over, providing your job completes and exits.
+
+On clusters where you can run multiple jobs per node, running this will cause you problems 
+between your own jobs, because it would delete your second job's semaphore arrays too. In that 
+case, you have two options. You can limit your GAMESS jobs to only running one after the other:
+
+```
+# submit a job that will only start after job 12345 ends
+qsub -hold_jid=12345
+```
+
+Or you can request exclusive use of the node you are using so nothing else runs there (if you do 
+this, do use all of the cores available in the node if possible so they are not wasted):
+
+``` 
+# request exclusive use of this node so that no other jobs run on it
+#$ -ac exclusive
 ```
 
 #### Troubleshooting GAMESS