Hodor HPC Cluster Head Node L O N

  • Slides: 15
Download presentation
Hodor HPC Cluster Head Node L O N M N G Comp Node Parallel

Hodor HPC Cluster Head Node L O N M N G Comp Node Parallel Storage H P N

Hodor HPC Cluster Head Node Comp Node L O N M N G Comp

Hodor HPC Cluster Head Node Comp Node L O N M N G Comp Node Parallel Storage H P N Head Node • 2 CPUs – 8 Cores - Sandy. Bridge • 64 GB RAM • Module Files • Open MPI, Intel MPI • GCC, Intel COMPILERS • SLURM Job Scheduler

Hodor HPC Cluster Head Node Comp Node L O N M N G Comp

Hodor HPC Cluster Head Node Comp Node L O N M N G Comp Node Parallel Storage H P N Head Node • Must Request an Account • Authentication • Campus Connection • File Transfer • SCP • SFTP • Globus • Compile • Very Short Test Runs • Do not USE mpirun / mpiexec

Hodor HPC Cluster Head Node Comp Node L O N M N G Comp

Hodor HPC Cluster Head Node Comp Node L O N M N G Comp Node H P N Head Node • SSH Keys Are Allowed • Must Append new keys • Replacing keys will break your account Comp Node Parallel Storage • Must be on-campus to connect • Or use Campus VPN • Incorrect password attempts will block user access.

Hodor HPC Cluster Head Node Comp Node L O N M N G Comp

Hodor HPC Cluster Head Node Comp Node L O N M N G Comp Node Parallel Storage H P N Compute Nodes x 32 • 2 CPUs – 4 Cores – Sandy. Bridge • 16 Nvidia Tesla K 20 m GPUs • 16 Xeon Phi 31 P 1 s Coprocessors • 64 GB RAM • Module Files • Use “srun” to get Bash Shell

Shared File Space /home/user. name /share/apps /cm/shared Head Node Comp Node L O N

Shared File Space /home/user. name /share/apps /cm/shared Head Node Comp Node L O N M N G Comp Node Parallel Storage H P N Not Shared /tmp Heavy I/O should be done in: /tmp Must be copied before end of job.

SLURM Queue Information Commands squeue sinfo qstat –a qstat –n qstat –f

SLURM Queue Information Commands squeue sinfo qstat –a qstat –n qstat –f

SLURM Job Submission Commands srun – for interactive - use of “screen” sbatch -

SLURM Job Submission Commands srun – for interactive - use of “screen” sbatch - for batch runs

SLURM Job Submission Commands Examples: /share/apps/slurm/examples sbatch somescript. sh srun –N 1 –n --pty

SLURM Job Submission Commands Examples: /share/apps/slurm/examples sbatch somescript. sh srun –N 1 –n --pty bash

Important sbatch arguments #SBATCH –N 8 #SBATCH --ntasks-per-node=8 #SBATCH –t 00: 10: 00 #SBATCH

Important sbatch arguments #SBATCH –N 8 #SBATCH --ntasks-per-node=8 #SBATCH –t 00: 10: 00 #SBATCH –o. /out_test. txt #SBATCh –e. /err_test. txt

Important sbatch environment variables $SLURM_SUBMIT_DIR Path from current dir when sbatch was invoked. $SLURM_NTASKS

Important sbatch environment variables $SLURM_SUBMIT_DIR Path from current dir when sbatch was invoked. $SLURM_NTASKS Total number of tasks in your job $SLURM_JOB_ID ID number identifying your SLURM job.

SLURM Job Delete Commands scancel #### qdel ####

SLURM Job Delete Commands scancel #### qdel ####

Modulefile Commands module avail module load module list

Modulefile Commands module avail module load module list

Hodor Help Info Website: www. crc. und. edu Tutorials & Desktop Software > Linux

Hodor Help Info Website: www. crc. und. edu Tutorials & Desktop Software > Linux HPC Cluster (Hodor) Hodor does not have a job submission queue called “test”

Get Hodor Account Send email to: Aaron. Bergstrom@und. edu

Get Hodor Account Send email to: Aaron. Bergstrom@und. edu