Manual

Important

Please note that from outside our network, including the voucher network, you first have to connect to hydra, i.e., ssh hydra, before connecting to the compute servers!

The servers fall into two distinct categories: interactive and batch. Interactive servers (compsrv01 .. compsrv13) can be accessed directly via ssh, i.e., enter

ssh compsrv01

and run your programs directly from the command line. You may also use the general name compsrv to randomly choose one of the compute servers:

ssh compsrv

(For ssh configuration, see also below.)

Please also refer to Hardware and the status of all compute servers and choose one suited for your job, e.g. with little load or memory usage.

Batch servers may only be accessed via the Batch System.

Filesystems

Home Directories

On the compute servers the home directories are different from the standard home directories, i.e., available when using the standard desktop computers. Hence, you first have to copy your programs or data to the clusterhome directory.

Note

The file system quota is 1TB per user. If you need more disk space, either use /scratch or ask for a separate disk space.

On the desktop computers, this is accessible via

/clusterhome/<username>

If the corresponding directory does not yet exist, e.g., you see an error message, please log into hydra, which triggers the automatic creation of the directory.

Important

As /clusterhome is for computational data, there is no long term backup available! Snapshots will be available for at most four weeks!

If you need to access your standard home directory from the compute servers, you can do so via

/mishome/<username>

Since access to the standard home directory may trigger a problem with the access to our file servers, please avoid this if you can. It is safer to do all copy operations between both directories from the desktop computers (which also have /clusterhome mounted).

To get the permissions to access your standard home directory on the compute servers, you first have to run

kinit

Note

/clusterhome has relaxed security settings to provide better performance on the compute servers. Hence, do NOT store sensitive data on /clusterhome.

Scratch space

For temporary files, e.g., created by jobs during computation and not intended for backup, a special filesystem is available:

/scratch

/scratch is a distributed, parallel filesystem using hard drives of several compute nodes.

It can be used like /tmp on other Unix systems.

Please note, that no backup is made of the data on /scratch.

By default, all data is stored thrice in the file system, i.e., three servers will hold a copy thereby providing some fault tolarance should one server be down.

Software

Standard Software

Most of the software on the compute cluster is accessible through the standard system paths, e.g. /usr/bin. This includes

Compilers and Interpreters

Editors

Tools

Applications

GCC (C, C++, Fortran)

Emacs

GNU Make

Octave

Java

NEdit

CMake

GNUPlot

Python

JEdit

SCons

SCILab

Perl

Eclipse

R

Ruby

GAP

OCAML

Axiom

Erlang

Special Software

In addition to the above, the following software can be found in /opt/local.

  • Intel Composer, e.g. C, C++ and Fortran

  • Matlab

  • Maple

  • Mathematica

  • SageMath

  • Molpro

  • Gromacs

  • Magma

  • …

Commercial Software

Should you need software not available on the compute servers, please take a look at the list of commercial software handled by the Software Licensing Group of the MPG available here and contact us so we can manage availability also at our institute.

Magma

The license for Magma is bound to the following compute server: compsrv10. It will not work on a different server!

Modules

Instead of starting the above programs from the sub directories in /opt/local a environment module system is available (for most packages), which allows you to make it available in you standard environment. This will set the correct variables in you shell environment for running the programs.

To see, which modules are available, use

module avail

You may load a specific module via

module load ...

e.g.

module load compxe
module load matlab
module load mathematica

for the Intel Composer XE, Matlab and Mathematica. You may put these statements into you shell startup files, e.g. .bashrc or .cshrc to have the modules automatically loaded upon login.

Similarly, the software may be remove from your environment via

module unload compxe
module unload matlab
module unload mathematica

The list of currently loaded modules is printed by calling

module list

SSH Configuration

Since the different compute servers are accessed via the single name compsrv, ssh will report potential man-in-the-middle-attacks. To prevent this, you should add the following lines to ~/.ssh/config (your SSH configuration):

Host compsrv
    StrictHostKeyChecking    no
    LogLevel                 QUIET
    UserKnownHostsFile       /dev/null

To automatically add the -X option for enabling graphical applications each time you log in to a compute server, you can also add:

Host * 
    ForwardX11Trusted  yes
    ForwardX11         yes

You can use the command

modify-ssh-compsrv

in a terminal application to perform the above modifications to your SSH configuration.