Using Stata on the CCS cluster

• If you don't already have one, obtain a CCS account (username). Let Keith know, so he can add you to the econ workgroup on the server.


• Write your Stata do-file (mydofile.do). Example:

log using mydofile.log, text replace
set mem 300m
sysuse auto
generate weightsq = weight^2
regress mpg weight weightsq foreign
log close
exit




• Write a script for the CCS queueing system (myscript.sh). Example:

#PBS -q econ_long
#PBS -N mydofile
#PBS -d /scratch03/username
#PBS -j oe
#PBS -o stdlog.log
#PBS -l nodes=1:ppn=2,walltime=5:00:00,mem=1gb
#PBS -M username@tulane.edu
#PBS -m abe

module load intel
module load stata

echo Start Job
cd /scratch03/username
date
pwd
/usr/local/opt/stata/stata-mp -b do mydofile
echo End Job


This script requests that the contents be added to the econ_long queue. It will let the script run for up to 5 hours, on 2 processors, on 1 node, using 1 GB of memory, and send out an email when it begins, finishes, or aborts.



• Upload your do-file and queue script to sphynx.ccs.tulane.edu using secure FTP (SFTP).

To do this, you need an FTP program, like Filezilla. The server address is sphynx.ccs.tulane.edu. Specify SFTP and use your CCS username and password. Once you login, change the directory to /scratch03/username. Then, upload mydofile.do and myscript.sh to that directory.



• Login to spynhx.ccs.tulane.edu via SSH

To do this from Windows, you need an SSH terminal, like Putty. To do this from Mac OS, you can use the built-in Terminal. The server address is sphynx.ccs.tulane.edu. Specify SSH and use your CCS username and password. For example, in Terminal:

username$ssh sphynx.ccs.tulane.edu
username@sphynx.ccs.tulane.edu's password:
username@sphynx>




• Change directory to /scratch03/username.

username@login-02-01>cd /scratch03/username




• Add your job to the PBS queue.

username@login-02-01>qsub myscript.sh
jobid




• Check the status of your job.

username@login-02-01>qstat




• Check the status of the nodes.

username@login-02-01>qnodes




• Kill your job if necessary.

username@login-02-01>qdel jobid




• Other information about the queuing system can be found here.