Differences

This shows you the differences between two versions of the page.

--- compiled_matlab_on_igrida [2013/07/20 11:26] – created samer
+++ compiled_matlab_on_igrida [2014/03/02 17:44] (current) – [Preparing the Scripts] samer
@@ Line 1: / Line 1: @@
-===test===
+======Compile your MATLAB scripts and submit as a batch job=======
+This tutorial details the procedure to compile your MATLAB script and launch it as an OAR batch job on IGRIDA.
+=====Logging In=====
+Start by connecting the IGRIDA frontend using SSH:
+<code bash>
+mymachine%ssh mylogin@igrida-oar-frontend
+</code>
+When at the IGRIDA front-end, start an interactive job:
+<code bash>
+igrida-oar-frontend%oarsub -I
+</code>
+=====Preparing the Scripts=====
+Let us assume that your main MATLAB function is called ''main_foo''. This function has two input parameters and calls another function called ''sec_foo''. These two functions are stored in your home directory ''/udd/mylogin''. However, any output must me stored (for example in .mat files) on ''/temp_dd/igrida-fs1/mylogin/'' and never on your NFS home directory.
+<file bash main_foo.m>
+function main_foo(x,y)
+a=x+y;
+c=sec_foo(a,x);
+matFilename = sprintf('/temp_dd/igrida-fs1/mylogin/results.mat');
+save(matFilename,'a','b','c');
+</file>
+<file bash sec_foo.m>
+function c = sec_foo(a,x)
+c=a*x;
+</file>
+In order to compile your scripts, you can use the ''mcc'' compiler. Note that you should limit MATLAB to a single computational thread (by default, MATLAB makes use of the multithreading capabilities of the computer on which it is running). All your MATLAB scripts should be given as arguments starting with ''main_foo.m''. The syntax of the ''mcc'' command is given hereafter while in the interactive OAR session:
+<code bash>
+igrida01-01%module load matlab
+igrida01-01%/soft/matlab_hd/R2012b/bin/mcc -R -singleCompThread -m main_foo.m sec_foo.m
+</code>
+Now, you have got an executable called main_foo that you test on the command line by typing:
+<code bash>
+igrida01-01%./main_foo 10 20
+</code>
+You can now exit the interactive OAR session and get back to the frontend.
+=====Preparing the OAR jobs=====
+Let us suppose that you need to run hundreds of simulations of your MATLAB compiled script. You can choose to modify your MATLAB scripts and add iterating loops (then recompile). Then, you can follow this [[matlab_script_on_igrida|tutorial]] to run your job. However, with this method, you launch a single OAR job and do not take full advantage of the computing facilities on IGRIDA.
+In order to run multiple jobs, you can write a launcher script as in ''oar_launcher.sh''. This script outputs a set of jobs in separated script files called job*.sh. In each job file, you call the compiled main_foo with incrementing parameters. Note that you may need to change the walltime and the number of cores. The launcher ends by submitting all the created jobs.
+<file bash oar_launcher.sh>
+#!/bin/bash
+for i in {1..10};
+do
+	for j in {1..10};
+	do
+		echo "#!/bin/bash " > job$i$j.sh
+		echo "source /etc/profile.d/modules.sh" >> job$i$j.sh
+		echo "module load matlab" >> job$i$j.sh
+		echo "#OAR -l core=10,walltime=3:00:00" >> job$i$j.sh
+		echo "#OAR -O /temp_dd/igrida-fs1/mylogin/log/job.%jobid%.output" >> job$i$j.sh
+		echo "#OAR -E /temp_dd/igrida-fs1/mylogin/log/job.%jobid%.error" >> job$i$j.sh
+		echo 'echo "My job was ran on these nodes:"' >> job$i$j.sh
+		echo 'cat $OAR_NODEFILE' >> job$i$j.sh
+		echo "#Setup MCR cache directory locally" >> job$i$j.sh
+		echo 'export MCR_CACHE_ROOT=/tmp/mcr_cache_${USER}_OAR_JOBID_${OAR_JOBID}' >> job$i$j.sh
+		echo 'mkdir -p $MCR_CACHE_ROOT' >> job$i$j.sh
+		echo "#cd to your execution directory first" >> job$i$j.sh
+		echo "cd /udd/mylogin/" >> job$i$j.sh
+		echo "./main_foo $i $j" >> job$i$j.sh
+		echo "#Remove temporary MCR cache directory" >> job$i$j.sh
+		echo '/bin/rm -rf $MCR_CACHE_ROOT' >> job$i$j.sh
+ 	done;
+done;
+for k in job*;
+do
+    chmod +x $k;
+    oarsub -S $k;
+done;
+</file>
+Do not forget to create your log directory:
+<code bash>
+igrida-oar-frontend%mkdir -p /temp_dd/igrida-fs1/mylogin/log/
+</code>
+Prepare you OAR launch script to be executed:
+<code bash>
+igrida-oar-frontend%chmod +x oar-launcher.sh
+</code>
+=====Launching the Job=====
+You can now launch your jobs by simply calling your launcher script as in the following:
+<code bash>
+igrida-oar-frontend%./oar-launcher.sh
+[ADMISSION RULE] Modify resource description with type constraints
+[ADMISSION RULE] Job walltime greater than 4 hours, adding property duration_weight > 2 (large job).
+Generate a job key...
+OAR_JOB_ID=666666
+[ADMISSION RULE] Modify resource description with type constraints
+[ADMISSION RULE] Job walltime greater than 4 hours, adding property duration_weight > 2 (large job).
+Generate a job key...
+OAR_JOB_ID=777777
+</code>
+Step back and watch your jobs being successively created then executed.
+=====Verifying the Output=====
+If you need to follow the execution of your job, you can check the status of all your jobs:
+<code bash>
+igrida-oar-frontend%oarstat -f | grep mylogin
+</code>
+and verify the output or the errors generated using a specific job ID:
+<code bash>
+igrida-oar-frontend%more /temp_dd/igrida-fs1/mylogin/log/job.jobID.output
+igrida-oar-frontend%more /temp_dd/igrida-fs1/mylogin/log/job.jobID.error
+</code>