University of Hamburg - to research, to teach, to educate and form, to homepage
RRZ
Regional Computer
Center (RRZ)

Photo: Thomas Orgis

quick start

Contents

The HPC paradox

Quick Start begins with a generic topic: The HPC paradox. The rest of this page describes pecularities of the Hummel-2 cluster.

We would like to start with mentioning the HPC paradox that every HPC user is faced with.

  • An HPC cluster is not automatically faster than your personal computer.

Please contact the HPC team before you launch production jobs if you feel uncertain about these issues.

Login

The computer from which you want to access a login gateway must be in the network of Hamburg University. From outside the university VPN must be employed.

There are 2 login gateways. Their IP addresses are:

hummel3.rrz.uni-hamburg.de
hummel4.rrz.uni-hamburg.de

Login is possible with the Secure Shell using public key authentication.

Note that you must log into a front-end node (see below) to start working. It is possible to transfer data via the login gateways but there is only a minmal set of software installed.

For X11 forwarding see the page on X11 forwarding.

Hint for MacOS users: If you get locale error messages from the Module setup try to follow the recommendations from https://blog.remibergsma.com/2012/07/10/setting-locales-correctly-on-mac-osx-terminal-application/.

Front-end nodes

A front-end node can be reached from a login gateway via:

bxy1234@hummel2-front1:~$ ssh front1
bxy1234@hummel2-front1:~$ ssh front2

The two ssh commands can be combined into a single one:

$ ssh -t bxy1234@hummel3.rrz.uni-hamburg.de ssh front1

The X11 forwarding option -X must be given to both ssh commands (recall that -X should only be used when needed):

$ ssh -X -t bxy1234@hummel3.rrz.uni-hamburg.de ssh -X front1

Operating system

The operating system is Debian GNU/Linux.

Configuration files

Initially the home directory $HOME is almost empty, i.e. there is a .ssh directory but no other configuration file like .profile or .bashrc.

Working with data

Data handling becomes an increasingly important topic. Please read also:

Summary

  • The home directory ($HOME) should only/mainly be used for files that you create/edit yourself.
  • The User-installed-SoftWare directory ($USW) should be used for installing software packages downloaded form the internet. $USW should als be used to unpack containers.
  • Data should be kept in the /beegfs ($BEEGFS) or /nfs/ssdX.Y ($SSD) file systems, see Which file system to use when?.
  • In batch jobs /tmp and /dev/shm are virtual file systems that are both kept in memory and are private to the batch job.

Disk quotas

Disk quotas are enabled on all filesystems. Disk usage and accessibility can be checked with the RRZ tool

Software

The operating system images of Hummel-2 contain only minimal software. To get more functionality switch to a pkrsrc Module. When you start on Hummel-2 it is a good idea to switch to a pkrsrc Module because this provides software that might be missing:

module switch env env/YYYYQQ-gcc-openmpi

Please install binary software packages (like Python packages) in your $USW directory.

More information on software is given on the page

Compilers

Installed compilers are listed on the page

Running batch jobs

Important points are:

  • The smallest scheduling unit is 8 cores or 1 GPU. In particular, the smallest CPU job size is: 8 cores, 32 GB RAM.
  • Single-core jobs are not allowed.
  • Very short (minutes) and very long (days) running jobs should be avoided.
  • Specifying --account at job submisson might be necessary.
  • Submit all jobs with
          #SBATCH --export=NONE
    and
          source /sw/batch/init.sh
  • Launch MPI programs with mpirun (srun does not work).
  • Check your batch job limits with rrz-batch-limits.
  • Check resource usage of your batch jobs with rrz-batch-jobreport.

Please read also: