How to use Scalasca on CRESCO

For using Scalasca on CRESCO, users have to load the appropriate environment by typing in at the prompt the command scalasca-2.3.1. Detailed information can be found in the /doc directory as suggested by the command itself. Once loaded the Scalasca environment, a cursory look at the main features can be done by giving the command scalasca -h.

The basic steps in using Scalasca for analyzing a program involve three phases: program instrumentation, measurement execution and analysis report:

After successful execution of the job analysis a summary report file is created within a measurement directory called scorep_*. The summary analysis report can be postprocessed by typing the command:

Scalasca comes with the GUI CUBE which allows the programmer to analyse the main metrics of the executable for eventually proceeding with further investigations.

It is possible to submit a Scalasca job to CRESCO in batch mode. This is accomplished by following the simple procedure described below.

• Step 0: Optional

  If the job will produce a huge amount of data or I/O, it should be better to use the ENEA GRID Parallel File System (GPFS).

  cd ~/PFS/porq1_1M
mkdir myfolder
cd myfolder

• Step 1: Load Scalasca

  Load Scalasca environment by typing

  scalasca-2.3.1

• Step 2: Instrument the code

  scorep mpicc myprog.c -o myprog

• Step 3: Submit to LSF

  Use a submition script from inside the Scalasca environment

  bsub -o myoutput -e myerror -n #cores -q quename ./submit.sh

  Where the script looks like:

  #!/bin/sh HOSTFILE=$LSB_DJOB_HOSTFILE PROCS=`cat $LSB_DJOB_HOSTFILE | wc -l` scalasca -analyze \ mpirun "-np $PROCS" \ "--mca plm_rsh_agent blaunch.sh" \ "--hostfile $HOSTFILE" \ -- ./myprog

  Pay attention to double quote the strings denoting mpirun parameters. Often it is necessary to put a double hyphen "--" before the target executable. More details can be found in the Scalasca User Guide. Also, it is recommended to read the OPEN_ISSUES file.