Picongpu: Spack Install on Barkla: Namespace Unknown

[cbontoiu]

Hello,

I recently got access to a cluster and someone who has the superuser power on it installed for me PIConGPU using gcc 5.5.0

I activated the Spack environment but there is some error when trying to load PIConGPU, namely "Unknown namespace: hzdr.crp". Do you have an idea of what can it be, please?
On the other hand the spack info picongpu does not work.
Thank you!

[sbastrakov]

Hello @cbontoiu . A small question: how did that other person install PIConGPU, also using spack or from source? I am guessing from spack, but just to be sure.

[ax3l]

I could imagine that the admin installed PIConGPU with a spack and provides you with modules that spack created.

Did you try module avail and if it lists either PIConGPU or its dependencies as modules?

[cbontoiu]

Thank you for your help. In the end the admin gave me the possibility to install spack and then PIConGPU and the installation went smoothly with the latest version and [email protected]. Now comes the most difficult part of launching a test case with Slurm. For now I see the our cluser has "There are three GPU nodes (i.e., gpu01 powered with 4 Nvidia Tesla P100 GPUs and gpu02/gpu03 powered with 4 Nvidia Tesla V100 GPUs) in “gpu” partition, which can used with the Slurm job scheduler." so I am pretty excited to give it a try for the first time on the cluster.
Regards.
Cristian

[PrometheusPi]

@cbontoiu If you have a look at our already difined cluster configurations where there are various example SLURM *.tpl files that you could use to derive your submit script.
E.g.:

• https://github.com/ComputationalRadiationPhysics/picongpu/blob/dev/etc/picongpu/hemera-hzdr/gpu.tpl
• https://github.com/ComputationalRadiationPhysics/picongpu/blob/dev/etc/picongpu/taurus-tud/V100.tpl
  
  
  
  • https://github.com/ComputationalRadiationPhysics/picongpu/blob/dev/etc/picongpu/pizdaint-cscs/normal.tpl
  
  
  • https://github.com/ComputationalRadiationPhysics/picongpu/blob/dev/etc/picongpu/lawrencium-lbnl/k20.tpl
  
  

Please be aware that the first two (hemera/HZDR and taurus/ZIH) require quite up-to-date SLURM versions while the last two (pizdaint/CSCS and lawrencium/LBNL) uses an older version of SLURM.
The main difference is how you set your working directory: either by --chdir(new) or --workdir(old).

I hope this helps you to derive your own *.tpl file. We would be happy to include it in our list.

If you have any questions about defining your *.tpl file, feel free to ask.