Frequently asked questions about Aurora

Author: Joachim Hein (LUNARC)

Node icon

I have forgotten my login password for Aurora

To receive a new password, please visit the password self service portal; identify yourself with your email address and the mobile phone you have registered you have registered in SUPR. If this fails, phone LUNARC support on 0462224454 (+46462224454 outside Sweden) from your registered mobile phone. Additional questions will be asked to verify your identity.

I can not login after installing a pocket pass token

Many users have issues because of missing the activation step. To activate your token, connect to the self-service portal, go into "tokens", choose "activate" and follow the instructions.

I have persistent problems connecting after beging prompted for my OTP

If you have persistent problems with connecting to LUNARC service after being prompted for my password and pocket pass password, it my be that your pocket pass token has expired. Please check the expiration status of your pocket pass token and create a new token if the old token has expired.

Could your send my one time password for pocket pass activation to my new mobile phone number?

Please enter the new phone number into SUPR and raise a support request to update our internal database for this change.

I accidentally deleted or modified a file

For the home space on Aurora, we have snapshots enabled. If you type ls .snapshots at the commandline in the current directory you get to see a number of diretories named like @-2020.12.11-00.00.35, which stands for the date and time of the snapshot. Change into the directory specifying a point of time before the accident, e.g.:

cd .snapshots/@-2020.12.11-00.00.35

and you get presented with the state of the directory and it's sub-directories at that time-stamp. You can copy the required contents out of the snapshot directory into your current directory.

What is the maximum job time on Aurora?

We allow batch jobs to ask for up to 168 hours, which is 7 days.

Could you please install software for me

Many packages that were installed in standard locations have now been moved to the module system. Please search for the package using module spider before contacting the helpdesk.

I can not find my software package in the module system

On Aurora a hierarchical module naming scheme is deployed. This is explained in a special software guide. Please contact LUNARC support if you cannot locate your package with the module spider command.

Loading a module unloads pre-requisites from a previously loaded module

Many software modules on Aurora have complex pre-requisites. If you load multiple modules is can happen that they try to load conflicting modules as pre-requisites. In this case the module loaded last unloads conflicting pre-requisistes from earlier loaded modules. If that happens, the earlier loaded modules can become disfunctional and there is no guaranty that they still work.

You have the follwing options to resolve the issues:

  • Have a look, whether there are different versions of the modules installed that have common pre-requisites. Acutally in many cases we have the same version of the software installed with different pre-requisites to help this issue.

  • Have a terminal dedicated to each of the modules. This should work if the modules do not really interact with each other, but e.g. one module creates files which the other module reads. Consider using the Lunarc HPC desktop which easily allows having multiple terminals within a single session

  • If this does not work or does not work satisfactory, contact the LUNARC helpdesk and discuss your situation.

How do I use a Lund University project

When using a Lund University project (LU project), you have to specify the project name, the partition and in some cases a reservation. This is explained in our batch system guide.

I want to use the Aurora GPU nodes

To access the Aurora GPU nodes, you need to be a member of an LU local project. Please review our batch system guide for the required modification to your submission script. The software guide provides an overview on Compiler, Cuda and MPI support for the GPU nodes.

Can I prevent my job from restarting in case of a node failure

By default the Slurm job scheduler restarts jobs in case of e.g. a failing node. This is not always desirable. Adding a line

#SBATCH --no-requeue

to the header portion of your job script prevents this behaviour.

I need LAPACK and BLAS - they used to be in /usr/lib64

Using an optimised BLAS library is important for achieving good performance. The libraries in /usr/lib64 are typically not highly optimised. We currently support OpenBLAS within the foss toolchain and MKL in the intel, iomkl and gimkl toolchains. Both libraries show excellent performance when e.g. used to build HPL (high performance linpack).

When using OpenBLAS, please be aware that you might need to control the number of threads spawned.

The Aurora modules providing MKL set the environment variable MKLROOT to assist the compiler in locating the library on the system. To link your application we suggest consulting the IntelĀ® Math Kernel Library Link Line Advisor for the arguments needed.

I want to install an R-extension

LUNARC provides R on its systems. The R installations have many R extensions build in. If you require an R extension that is not provided by the installation, you can easily install it in your home space. We describe this in your R guide.

My MPI code using OpenBLAS does not perform

OpenBLAS as installed in the foss toolchains is compiled with thread support. When OpenBLAS is used on Aurora, it will typically figure how many cores you are allocated on that node and spawn as many threads. If this is not desirable (e.g. pure MPI code) you need to control the number of threads spawned by setting the OMP_NUM_THREADS variable. For example for an MPI code running as many tasks as cores requested this is typically be set to:


I like to run VASP on Aurora but it doesn't work

If you like to use VASP on Aurora, please ensure that you are registed on a license with the VASP developers in Vienna. To comply with the terms of our handling license, we have to confirm the validity of your license with the VASP developers before granting access. That will take normally take a few days but can take weeks under bad circumstances. Being named in a SUPR VASP group does not imply you are entitled to use VASP.

After registerering with the VASP team, contact the LUNARC helpdesk and provide the following information:

  • Your LUNARC userid
  • Name and Department of the license holder (typically your supervisor/research group leader)
  • The number of the license
  • Your name as stated on the license
  • Whether you require VASP 4 or VASP 5 access

In order to be able to continue the provision of VASP to our ligitimate VASP users, we will not make any exception.

The VASP5 executables on Aurora are compiled using Intel MPI. They need to be started with srun. The VASP6 executable is build with OpenMPI. These executables need to be started with mpirun. Refer to the sample section of our batch system guide for a more comprehensive discussion.

My MPI application doesn't launch

On Aurora we support two MPI libraries: OpenMPI and Intel MPI. Depending on the toolchain utilised to compile your application different job launchers are required. Executables build with OpenMPI need to be launched with mpirun, while applications build with Intel MPI need to be started with srun.

Visual Studio Code (code) doesn't work anymore

Microsofts code editor Visual Studio Code in its latest version uses a runtime library that is no longer suppoerted by CentOS 7. To fix this load the foss module before running Visual Studio Code

module load foss

Access to MAX-IV files from Aurora

If your MAX-IV project (i.e., "proposal" in MAX-IV terms) files are not already accessible on Aurora on /projects/, then send a ticket to LUNARC via SUPR to request it. Mention the project name and, if possible, the UNIX Group name and GID.

Note that when MAX-IV project files are made accessible on Aurora, access is only possible from frontends, i.e., they are not accessible from compute nodes. Before starting a job, please copy the files from /projects/ to your home directory, then modify the job to use the copies in your home directory before submitting it.