Queue

eval15 is a cpu intensive program. Different experiments or even different boxfiles can be processed independently. The following set of commands use the gridengine queue system to process a set of boxfiles.

QueueSetup
QueueStart
QueueWait
QueueClean

QueueSetup

Syntax: queuesetup -m experiment/first/rest/set/thread [-s/-ns] [-c/-nc]
Default: queuesetup -m set -s -c

- -m experiment
  eval15 integration for all boxfiles in one experiment. The average shifts will propagate through the sequeuence of boxfiles.
- -m first
  eval15 integration on all first boxfiles of every experiment. When the average shifts for these boxfiles have been determined, you can use queuesetup -m rest to run eval15 on the remaining boxfiles. The saved shifts of the first boxfile will be used as initial shifts of the remaining boxfiles of the corresponding experiment.
- -m rest
  Independent eval15 run on on boxfiles. The initial shifts used for every boxfile are the results of the average shifts of a first boxfile in every experiment (i.e. after queuesetup -m first)
- -m set
  eval15 integration per boxfile. One job for each boxfile. If the number of boxfiles is large, consider -m thread.
  If, by coincidence, the average shifts of a specific boxfile sNNfMMM.shoe have been saved, they will be used if sNNfMMM+1 is started.
- -m thread
  eval15 integration per boxfile. The number of jobs is set with the -t option (default=48). Every job will process a number of sets. This option is useful with large numbers of boxfiles.
  If, by coincidence, the average shifts of a specific boxfile sNNfMMM.shoe have been saved, they will be used if sNNfMMM+1 is started.
A list of experiments/boxfiles (joblist) is created. This list is sorted on boxfile size or experiment size (largest one first), unless the commandline argument -ns has been specified.
A script joba is created. This script will be invoked by the gridengine system using an environment variable SGE_TASK_ID. This variable specifies the n'th line in joblist. An eval15all run will be executed for that specific line (experiment or boxfile or list of boxfiles).
A script jobb is created.
- -m first
  Nothing will be done
- -m rest
  mergey on all *.y files
- otherwise
  mergey on all files from joblist
If -c has been specified, jobb will try to remove:
- for every zero-length error file joba.eJJJJ.K (with JJJJ the job id of joba and K the sequence number of the array job):
  - joba.eJJJJ.K (stderr of eval15 run)
  - joba.oJJJJ.K (stdout of eval15 run)
- If no non-zero errorfiles exist:
  - for every boxfile sMMfNNN.shoe:
    - sMMfNNN.log (eval15 text output)
    - sMMfNNN.y (eval15 reflection output)
  - joblist. The list of jobs.
  - joba jobb. The batch scripts.
jobb will create a file jobb.done once the yfiles are merged. The cleanup will be running after the creation of jobb.done.
A script jobc can be created by queuestart. This script can be used to wait for termination of jobb. See the wait argument of queuestart.

QueueStart

syntax: queuestart [wait]

submits the arrayjob joba to the gridengine queue eval.q. The number of lines in joblist is specified as argument to the qsub command to indicate the size of the arrayjob. The gridengine system will start joba for every entry in joblist.
submits job jobb to the gridengine queue eval.q. This job will only start running if all the processes of joba have finished.
queuestart ends here, unless the the optional argument wait has been given. In that case, a third job jobc is submitted to the queuesystem. This job will simply wait for completion of jobb and the terminal window will synchronize on jobc. A (premature) control-c in the terminal window will abort jobc and will not affect joba or jobb.
The jobs will be submitted to the queue eval.q. The environment variable queuename may be used to select a different queue. For example:
```
export queuename=main.q
queuestart
```
or
```
queuename=main.q queuestart
```

QueueWait

Use this command to monitor the progress of the batch process, started with queuestart. As long as jobb.done is not created, this command repeatedly invokes

qstat -q $queuename -s r | tail -n +3 | sort -k 6; qstat -s p

queuename defaults to eval.q

qstat -q $queuename -s r | tail -n +3 | sort -k 6
print all running (-s r) jobs of this user in the queue $queuename. Strip the first two lines of the output, and sort the jobslist on starttime.
qstat -s p
show pending jobs of this user

QueueClean

syntax: queuestart [-a]
When the batch process has finished, a lot of files are created.

joblist. The list of jobs.
joba jobb jobc. The batch scripts.
jobb.done. Finished signal of jobb
for every boxfile sMMfNNN.shoe:
- sMMfNNN.evc average shifts
- sMMfNNN.log eval15 text output
- sMMfNNN.y eval15 reflection output
- joba.oJJJJ.K stdout of eval15 run
- joba.eJJJJ.K stderr of eval15 run
JJJJ is the job id of joba. K is the sequence number of the array job.
jobb.oLLLL stdout of jobb
jobb.eLLLL stderr of jobb

If all processes terminated normally, joba.eJJJJ.K and jobb.eLLLL will be empty files, otherwise they will contain the error message describing the failure of the job.

queueclean removes all joba.o*, joba.e*, jobb.o* and jobb.e* files if the corresponding .e* files are empty. If an .e* file is not empty, the corresponding .o* file is also not removed.
If all .e* files are empty, queueclean -a also removes all s *.log and s *.y files (for every s in joblist).
And finally, if all .e* files are empty, queueclean removes joblist, joba, jobb, jobb.done and jobc

EVPY Suite Overview