IDL provides data analysis and visualization for small to medium size datasets. IDL does not operate in parallel and as such all individual analysis tasks must fit on a single node. That is, data to processed must fit within the RAM of a single node. Data that can be analyzed separately and do not require inter process communication can be spread across several nodes and processed independently. Time series data often has this property.
Using IDL on Blue Waters
Use of IDL on the login nodes is discouraged as is all data analyis. To use IDL on Blue Waters it is recommended that an interactive single node partition be obtained and the application run on that partition. One can also run IDL as a batch job on the compute nodes by selecting off screen buffering of the images and saving them to disk as they are created. See the example given below.
To use IDL interactively, load the module file "module load idl". Then start the command line interpreter by typing the command "idl". This will give you a command prompt. Type idl commands at the prompt. Plotting windows appear on you workstation or laptop if you are running an x windows session. Microsoft windows users will have to obtain some sort of X emulation for this to work.
Scripting IDL from compute nodes.
IDL can be used from compute nodes by creating a file with idl commands in it. For example place the following commands in a file called test.com.
myplot = plot(data,BUFFER=1)
myplot.save, "wiggle.png", HEIGHT=480,WIDTH=640
Notice the "BUFFER=1" argument to the plot command. This argument causes the plot command to not open a window for plotting and to plot off screen. Once this test file has been created one can call it in a pbs batch script in the following way.
#PBS -j oe
#PBS -N idltest
#PBS -l nodes=1:ppn=1
#PBS -l walltime=00:05:00
module load idl
aprun -B idl ./test.com
This pbs script runs the idl commands whyc create the wiggle.png file that contains the graphics output.