====== Altair HyperWorks ====== ===== Overview ===== ^Available on|Mesocomte cluster^ ^Installed version|11.0| ^Location on the cluster|/Softs/lumiere/Hyperworks| ^Web links|[[http://www.altairhyperworks.com/hwhelp/Altair/hw11.0/help/hwsolvers/hwsolvers.htm?run_options_for_radioss.htm|User guide]]| ^Licensing|Licensed to S.Roth | ===== Usage ===== To load the software just type: $ module load altair/HyperWorks/11.0 The launcher scripts are located in /Softs/HyperWorks/11/altair/scripts. When using ''modules'' this location is added to your ''PATH'' along with the license server information. Use ''$ALTAIR_HOME'' environment variable to access to the installation folder. ===== Interactive mode ===== Interactive mode is like starting graphical user interface for post/pre processing. You need to used a dedicated node for that: Connect to : ''mesoshared@univ-fcomte.fr'' using [[nx_client|NXClient]] To start HyperMesh for example, after loading module hyperworks with modules, juste type : $ hm ===== Batch mode with SGE ===== Connect to cluster to submit jobs: ''mesologin1.univ-fcomte.fr'' ==== Multi-threaded Execution (SMP) ==== A template example to run radioss in SMP machine: #!/bin/bash -l #$ -N scriptName ## script name #$ -pe openmp 8 ## request 8 threads #$ -l h_vmem=4G ## request memory 4g/core ## load the software module load altair/HyperWorks/11.0 radioss TWISBEAM_0000.rad -nthread $NSLOTS Required options: ^Option^ Description^ |-nthread|This is the number of parallel threads to use. It should be the same as the ''openmp'' that you specify| |-maxlen |The maximum memory in MB to us. This should be the same as the ''h_vmem'' that you specify| You can supply additional options for each script use the ''-help'' option to see additional arguments such as ''-analysis'' to do an analysis run. Also see the [[http://www.altairhyperworks.com/hwhelp/Altair/hw11.0/help/hwsolvers/hwsolvers.htm?run_options_for_radioss.htm|user guide]] for more detailed information. ===== MPI Execution ===== We need to start 2 jobs, for **compilation** and for **computation**. Compilation jobs will generate necessary files for the execution part. __**Compilation part** __ This phase needs a lot of memory depending on the number of SPMD parameters (**-np**) #!/bin/bash -l #$ -N radioss_compile ## script name #$ -q all.q ## queue to use #$ -l h_vmem=12G ## request memory 12g/core export OMPI_NUM_THREADS=1 ## load the software module load altair/HyperWorks/11.0 ##Number of SPMD part to generate SPMD=16 cp $ALTAIR_HOME/hwsolvers/bin/linux64/radflex_11_linux64 . s_11.0.240_linux64 -input hubys_2m3_0000.rad -np $SPMD This will generate 16 files for the computation part (.rst). **Submit your job** qsub radioss_compile.sge __**Computation part (MPI)** __ You MUST provide the same number of process as SPMD value in the previous script. 16 in this example This job will be executed in parallel using MPI using 16 cores #!/bin/bash -l #$ -N radioss_compute ## script name #$ -q parallel.q ## queue to use #$ -pe mpi 16 ## request 16 MPI cores #$ -l h_vmem=4G ## request memory 2g/core #$ -l radioss=1 ## for licence #$ -o compute.$JOB_ID.out #$ -V export OMPI_NUM_THREADS=1 ## load the software module load altair/HyperWorks/11.0 cp $ALTAIR_HOME/hwsolvers/bin/linux64/radflex_11_linux64 . echo "Running Engine at " `date` mpiexec.hydra -np $NSLOTS e_11.0.240_linux64_impi -input hubys_2m3_0001.rad echo "Finished at" `date` **Submission** qsub radioss_compute.sge ===== Tips ===== Note that computation job depends on compilation one. SGE allows to run jobs in sequence. We first run the compilation part with name radioss_compile qsub -N radioss_compile radioss_compile.sge And just after : qsub -hold_jid radioss_compile radioss_compute.sge