Running a job
this assumes you have a gaussian *.com input file ready to go
- jobs should be run on /scratch NOT in your home directory!
- cd /nfs/scratch/username where username is your login/user name, this command moves you to the scratch directory
- mkdir molecule make a new directory to work in, I make one directory per molecule, so replace molecule with a short name related to your molecule
- cd molecule change into the "working" directory
- cp /nfs/home/username/molecule.com molecule.com copy your *.com file called molecule.com from your home directory into your working directory on scratch
- make sure the top of your molecule.com file has the "%mem" and "%nprocshared" commands, this assigns the total memory required for the job and the number of cpu's to use, the numbers here will be good for most medium sized jobs. Here are the top few lines of an example input file:
%nprocshared=16 %mem=30GB %chk=molecule.chk # opt b3lyp/6-311g(d,p) geom=connectivity int=ultrafine
- make sure the "%chk" command has the SAME name as your *.com file, ie input file is molecule.com then checkpoint file is molecule.chk
- copy the highlighted text below for the "submission script" and put it into a file called g16run.sh. That is, in your terminal type vi g16run.sh a blank file will open, copy the text below with win-C(windows) or option-C(mac). Back in your file type "i" to get into insert mode and then paste (win-V or option-V). To save the file, go into command mode "esc" and type wq to write and quit the file
- the g16run.sh script (file) gives instructions to the batch queuing system about your job, you will need one of these files for every job you run, you can adjust the parameters accordingly, however this is a good script for most medium sized jobs.
- run time: the time set for this job is 1hr (time=1:00:00) if you have a big molecule you may need longer, first run the job with these parameters, if the job turns out to need more time, set-up the next one for 2hrs. You will quickly work out the minimum time you need. DO NOT be tempted to call a large amount of time, you job will go further back in the queue and this is wasteful of resources, resources cost money and deprive someone else of that resource.
- memory and cpu: as a beginner these can be hard to estimate, if your jobs are not completing with these parameters then ask me! We will look at increasing either or both of these parameters.
- submit the job to the que: sbatch g16run.sh
- check your job has entered in the que: vuw-myjobs
- made a mistake and need to cancel a job? run vuw-myjobs to get the jobID then use: scancel jobID
- you can look inside the *.log file while it is running! simply use vi, or the tail command is better
- when the job finishes you should have molecule.chk and molecule.log files in your directory (you may also have some other files, ignore them)
- look inside the *.log file, has it completed properly? if yes then these are *.chk and *.log files you want to keep, copy them back to your home directory cp molecule.log /nfs/home/username/molecule.log Of course you may have a molecule directory already set up in your home directory, in which case add that to the path in the above command
- when you are sure you have finished running jobs for that molecule, remove the molecule directory on scratch (make sure you have saved all the files you want to your home directory!)
- this should be enough to get you started, more info can be found on the raapoi documentaion wiki
- any problems e-mail me!
submission script
#!/bin/bash #SBATCH --job-name=molecule.com #SBATCH --cpus-per-task=16 #SBATCH --mem-per-cpu=2G #SBATCH --partition=parallel #SBATCH --time=1:00:00 #SBATCH --constraint=AVX #SBATCH -o /nfs/scratch/usrname/molecule.log #SBATCH -e /nfs/scratch/usrname/molecule.err module --quiet purge module load gaussian/g16 g16 molecule.com