Segmentation fault

[Balogh Szabolcs]

Hi again, 

So i can run Dirac perfectly on a VM created on our data server (scalable II / 2x16cores / 50Gb ram) and on my home computer (2x2686v4/128Gb ram) but we have segmentation fault on our cluster (4x2x40cores / 1Tb ram) which is driving me nuts. Its 4 nodes in a slurm cluster. No matter if i run the calculation on an individual node or through slurm from the masternode; same fault. 

I have found that it is maybe because of the 32bit installation of Dirac.

In the manual the bottleneck could be 16Gb/core. I dont know how this is counted but if i run a job with only 1node/40cores then we are talking about ~25Gb/core. 

Just realized while writing this that maybe i should run the jobs on at least 80 cores :)

That would be the aim anyway if i am moving on real complexes after evaluation.

The other related question is if Intel fixed the "64bit integer Intel MPI mpi_reduce"

problem? Would nice to know before putting much time in reconfigure the cluster.

Thank you again, 

Best regards,

Szabolcs

[Miroslav Iliaš]

Hello,

without knowing details,

with IntelMPI+MKL you can do the 64-bit integers installation of DIRAC (setup --int64 --mkl). Test it with ctest  in parallel. 

I have such buildup in https://github.com/miroi/open-collection/tree/master/theoretical_chemistry/software/dirac/buildup_on_servers/jinr_ru/hydra_jinr_ru/trunk_study_intelmpi_mkl_i8.

Miro

--
You received this message because you are subscribed to the Google Groups "dirac-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dirac-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/dirac-users/d73c0ea9-b183-4236-950c-c77c83f500dcn%40googlegroups.com.