Inside Collection (Course): An Introduction to High-Performance Computing (HPC)
Summary: In this module you will learn the fundamentals of editing, compiling and submitting codes on the CSCS Gele system.
For this tutorial you will be using the Ela front-end system for editing, compiling and running your codes. Your executables will be submitted on Gele which is the test system for CSCS's flagship machine Monte Rosa.
Please complete this section before attempting the remaining exercises. If you require any assistance, please do not hesitate to contact the available support staff.
From your local system, you can log into Ela (using your supplied username and password) with the following commands:
ssh -X username@ela.cscs.chWhen you log into the front-end node Ela, you need to set up your programming environment as follows:
module load gele
module load PrgEnv-cray
module load pbs
To compile source codes on Ela, please use one of the available compiler wrappers.
ftn -o foo foo.f90
cc -o foo foo.c
CC -o foo foo.cppCodes are executed by submitting a job script to a batch submission queue which in turn schedules the jobs to run on the compute nodes of Gele. To submit a job to Gele, please use the following submission script and submissions commands.
#! /bin/bash
#PBS -l mppwidth=8
#PBS -l walltime=00:10:00
#PBS -V
set -ex
cd $PBS_O_WORKDIR
aprun -n 8 ./foo.exe
exitIn this script we request 8 parallel tasks (mppwidth=8) for a maximum duration of 10 minutes (walltime=00:10:00). We ensure that our environment settings are exported to the compute nodes for execution (-V) and that our execution command is invoked within the same directory as that which the job script was submitted in.
To invoke an instance of the executable on the compute nodes, you are required to use the aprun command. You can run multiple instances of the same executable across multiple cores by including the -n parameter. In the sample batch script above, we have requested that our foo executable be duplicated on all 8 cores requested within the batch script resources.
You can modify most of the above settings if required. Please refer to the man pages (man qsub and man aprun) for further information.
To submit your job script to the batch queuing system use the following command:
qsub scriptWhen you submit your script successfully, the batch queuing system will issue you with a unique jobid for your submitted job. You can view the queuing and running status of your job by using the following commands:
qstat jobid
qstat -u usernameAfter termination of your job, you will receive stdout (standard output) and stderr (standard error) logs to your submission directory. These logs are suffixed with the id of your submitted job. Please review these logs carefully to determine whether you had successful or unsuccessful termination of your executable.
You are now in a position to attempt the "hands-on" practicals that complement the presentations. A number of practical exercises are included in this tutorial covering:
Before attempting the practicals you should copy across the exercise source templates to your home directory. To do this complete the following instructions:
A job queue maintained by the scheduler software for determining the order in which jobs are executed based on user priority, estimated execution time and available resources.
A compiler wrapper is a script that invokes the low-level compiler command with the required link libraries for a specific computing system and message-passing framework. Typically standard scientific and vendor-specific libraries are also included in the wrapper script.
The art of identifying and removing compile-time and runtime errors from a code.
Emacs is a feature-rich editor that has been in development since the mid-70s.
Please take a look at the emacs tutorial for more information on using the editor.
Ela is a 16 core front-end node which is used to edit and compile source code and submit jobs for execution on the compute nodes of Monte Rosa.
Gele is a 76 node dual-core system. It is the test system for the CSCS flagship machine Monte Rosa.
A script which describes the command line execution of an executable along with the core, memory and time resources required by the calculation.
A script the configures the search paths and link commands for a given software package or library.
The total number of cores requested by the user for the completion of the job.
The art of improving the memory performance and runtime performance of an executing code.
The Programming Environment is the collection of compiler and pre-built libraries that are specific to an individual compiler suite.
Monte Rosa is a Cray XT5 supercomputer with a peak performance of 140 TeraFlop/s. Its current specification is:
 |
Vi is a family of screen-oriented text editors which has been developed since 1976.
Please take a look at the Vi tutorial for more information on using the editor.
The total real time for completing a task. In a batch script, it refers to the maximum wall-clock time required by the job.
More about this module: Metadata | Downloads | Version History
This work is licensed by Tim Stitt Ph.D. under a Creative Commons Attribution License (CC-BY 3.0), and is an Open Educational Resource.
Last edited by Tim Stitt Ph.D. on Oct 5, 2009 12:50 pm -0500.