Fats : New Users : Basic Information for new users

NOTE: This is a template of an email sent to new users when their accounts on fats is setup. Please, make the proper adjustments when you read the text, e.g., userid should be read as your actual login name to fats.

Dear New_User_Name

Your account on fats has been setup. Your login name is userid. Your temporary password was sent to you on a separate email.

Please, keep this message for future reference.

All users of fats should abide by Columbia University Computer and Network Use Policy. I urge you to read it at:

http://www.columbia.edu/cu/policy/

Please, login to fats ASAP and change your password (using the passwd command). Please, let me know when you change your password. (But don't email the password to me!) I need to propagate it to the cluster nodes and to fats' file server, so that you can access them.

To login to fats please use ssh:

ssh fats

or

ssh -l userid fats.ldeo.columbia.edu

Your home directory at /home/fats/userid has a disk quota of 4GB (soft limit), with a hard limit of 4.5GB and a grace period of 7 days. In addition, you have a directory at /data[1,2]/fats/userid, with a disk quota of 20GB (soft limit), and a hard limit of 25GB and a grace period of 7 days. Moreover, you have a directory at /scratch[1,2]/fats/userid which is not limited by quota. Please, use disk space frugally, and remove big old files often.

Besides the regular Linux gcc, we have the PGI compilers, currently version 5.2-4. Hence, for compilation of serial code please use the the commands: pgf77, pgf90, pgcc, and pgCC, and for compilation of parallel code with MPI, please use the commands: mpif77, mpif90, mpicc, and mpiCC, for Fortran77, Fortran90, C, and C++, respectively.

PLEASE NOTE: When you login to fats, you are actually using the cluster's control workstation (a.k.a. front end). The purpose of this workstation is to provide to users a platform for code development. In addition, it is used as a file server and for system administration. In order to allow these three main tasks to be done properly, we ask users please to send executables to the cluster nodes, and not to run them on the control workstation. The cluster nodes, on the other hand, are reserved for running your jobs only.

Jobs are sent to the nodes only through the PBS queue system. The command used to submit these jobs is qsub. If your PBS script is called my_script.pbs, you can submit it by using:

qsub my_script.pbs

More information about PBS can be found in the manual pages of qsub (man qsub), and on the references therein.

You can monitor jobs on the PBS queue system using the utilities xpbs and xpbsmon. To make it easy, the script xp invokes both utilities.

Here is an example of a PBS script for a **serial** job, where the executable is called my_serial_program: 


    #!/bin/csh
    #PBS -o my_serial_job.o
    #PBS -e my_serial_job.e
    #PBS -l nodes=1:athlon:ppn=1
    #PBS -q production@fats.ldeo.columbia.edu
    #PBS -N my_serial_job
    cd $PBS_O_WORKDIR
    ./my_serial_program

And here is an example of a PBS script for a **parallel** job using MPI on 4 nodes (8 processors), where the executable is called my_mpi_program: 


    #!/bin/csh
    #PBS -o my_mpi_job.o
    #PBS -e my_mpi_job.e
    #PBS -l nodes=4:athlon:ppn=2
    #PBS -q production@fats.ldeo.columbia.edu
    #PBS -N my_mpi_job
    cd $PBS_O_WORKDIR
    mpirun -v -np 8 ./my_mpi_program

The examples above assume that the executables and the PBS scripts are located on the same directory. The examples also assume that you launch the qsub command from this directory.

On the examples above, both jobs are sent to the production queue. Code that is already debugged should be sent to the production queue. You can also use the checkout queue, if you just want a short run to check things out. Those jobs should use our nodes with dual Athlon (1.2GHz) processors, as specified on the PBS scripts above.

For code testing and debugging we have the development queue. The development queue uses our eight older/slower dual Pentium-III nodes. If you wish to use the development queue, please make the following changes on the scripts above:

athlon -> p3
production -> development

Please, note:

  1. Each node has 1GB of memory.
  2. The three file systems mentioned above, where you have directories, are exported/automounted and accessible from all nodes.
  3. Details about the queues (maximum resources, maximum time, etc) can be obtained from the utilities xpbs and xpbsmon, by highlighting the queue name and clicking on the button details.

If you are already familiar to PGI compilers, MPI compilation, PBS, etc, please forgive me for being so repetitive.

Don't hesitate to contact me if you need further assistance.

Welcome to fats!

Gus Correa
for fats' Sys. Admin.

Last updated on by Gus Correa.