Job description files
Submitting computing tasks to FGCI resembles submitting batch jobs to normal computing clusters. However, in the case of batch jobs the user just defines the commands to be executed, while in the case of grid usage the user must also define the (input) files that need to be transported with the job to the remote cluster and also the resulting (output)files that are returned when the job finishes.
In the case of ARC-middleware the grid jobs are defined using two files:
A job description file, that defines the resources needed (for example the required computing time, memory and number of processors) and the files that will be copied to and from the remote clusters. ARC can use xRSL or JSDL job description file formats.
A command file containing the commands that will be executed when the job is run in the remote cluster.
The command files are in most cases normal Linux command scripts. Linux scripting is not discussed in this guide. You can find more information about Linux scripting from the CSC workshop materials or via google.
The xRSL formatted job description files are text files that define the resources and files that the grid job needs. The file starts with an & sign, followed by a list of attribute-value pairs in the format:
Table 1. lists the most frequently used job description attributes. You can create xRSL formatted job description files with normal text editors.
Table 1. Most commonly used xRSL attributes
|count||Number of computing cores to be reserved.||(count=8)|
|cpuTime||Computing time requested. For a multi-processor job, this is a sum over all requested processors.||(cpuTime="6 hours, 20 minutes")|
|executable||Name of the command script file||(executable=runhello.sh)|
|inputfiles||Files that will be copied from the local computer to the remote cluster. The left column refers to the file name at the remote cluster and the right column refers to the file name in the local computer.||(inputfiles=("file1.txt" "file1.txt" ))|
|jobname||Name of the grid job||(jobname="hello_FGI")|
|memory||Memory requirement in megabytes||(memory="4000")|
|notify||Email will be sent to the given address at certain states of the job. E.g. When the job begins (b) or ends (e). Here the be signifies sending an email for both states.||(notify="be email@example.com")|
|outputfiles||Files that will be copied from the remote cluster when the results are retrieved. The left column refers to the file name at the remote cluster and the right column refers to the file name in the local computer.||(outputfiles=("out.txt" "out.txt" ))|
|runtimeenvironment||Required run time environment.||(runtimeenvironment="APPS/BIO/BOWTIE-2.0.0"|
|stderr||File for standard error||(stderr=std.err)|
|stdout||File for standard output||(stdout=std.out)|
Below is a short command script that is used as a simplified example of a grid command file. The job prints words "Hello FGCI" to the standard output and then writes the number of lines in files inputfile.txt and file2.txt to a new file called output.txt. The name of the command script in this example is runhello.sh.
#!/bin/sh echo "Hello FGCI" wc -l inputfile.txt file2.txt > output.txt exit
The runhello.sh script above can be executed in the FGCI environment using the following job-description file (called hello.xrsl):
&(executable=runhello.sh) (jobname=hello_FGCI) (stdout=std.out) (stderr=std.err) (cpuTime="1 hours") (memory="1000") (inputfiles= ("inputfile.txt" "file1.txt" ) ("file2.txt" "") ) (outputfiles= ("output.txt" "" ) )
The first line of the job description file defines that the script runhello.sh will be copied to the remote cluster and executed. The following lines define the name of the grid job (hello_FGCI) and the names of the standard output (std.out) and standard error (std.err) files.
The computing time (1 h) and memory (1000 MB) requirements of the job are defined in the fifth and the sixth rows. Defining these values is not mandatory but it is recommended that you do so. Setting up memory and time limits ensures that your job will be submitted to a cluster that has enough resources. Further, correctly set memory and time requirements ensure that in the remote cluster your job ends up in a queue that most effectively executes your job.
The definition (inputfiles= starts the region that lists the files that will be copied to the cluster executing the job. In addition to the actual input files of your job, this notation can also be used to copy program files like pre-compiled executables, source codes or program scripts to be used in the remote cluster.
The example above uses two alternative ways to define a file that will be copied to the remote cluster. The row:
("inputfile.txt" "file1.txt" )
defines that the file called file1.txt will be copied so that in the remote cluster the name of the file will be inputfile.txt. The next row:
defines that file file2.txt will be copied to the remote cluster. The same result could also be defined by a row such as:
The final closing bracket on a line by itself ends the input files defining regions.
A similar syntax is used to define the files that will be copied back from the remote cluster when the job results are retrieved. The output file defining regions starts with the notation (outputfiles= . In this example we will retrieve only one file, called output.txt. If you would like to retrieve all the files that are generated by the job execution directory in the remote cluster you can use the notation:
(outputfiles=("/" "" ))
When you define the output to be retrieved, it is good to remember that moving large files between the remote cluster and the local computer can take a long time. Thus, you should try to avoid unnecessary copying of large data-sets.