Slurm Wrapper Suite
The HAL Slurm Wrapper Suite is designed to help users use the HAL system easily and efficiently. The current version is "swsuite-v1.0", which includes
- Minimize the required input options.
- Consistent with the original "slurm" run-script format.
- Submits job to suitable partition based on the number of CPUs and/or GPUs needed.
- partition (required) : cpu_mini, cpun1, cpun2, cpun4, cpun8, cpun16. gpux1, gpux2, gpux3, gpux4, gpux8, gpux12, gpux16.
- [cpu_per_gpu] (optional) : 16 cpus (default), range from 16 cpus to 40 cpus.
- [time] (optional) : 24 hours (default), range from 1 hour to 72 hours (walltime).
- [singularity] (optional): specify a singularity container(name-only) to use from the $HAL_CONTAINER_REGISTRY
- [reservation] (optional): specify a reservation name, if any.
- [version] (optional): Display Slurm wrapper suite and Slurm versions.
To request a full node: 4 gpus, 160 cpus (→ 40*4 = 160 cpus) , 72 hours
or using a container image (dummy.sif) on a cpu only node with default of 24 hours
Note: In the second case we are using a singularity container image. To run a custom container, instead of using the default location of the container registry, you can set it to your own by first exporting the environment variable
Same as original slurm batch.
- RUN_SCRIPT (required) : Specify a batch script as input.
Within the run_script:
- partition (required) : cpu_mini, cpun1, cpun2, cpun4, cpun8, cpun16, gpux1, gpux2, gpux3, gpux4, gpux8, gpux12, gpux16.
- job_name (optional) : job name.
- output_file (optional) : output file name.
- error_file (optional) : error file name.
- cpu_per_gpu (optional) : 16 cpus (default), range from 16 cpus to 40 cpus.
- time (optional) : 24 hours (default), range from 1 hour to 72 hours (walltime).
- singularity (optional) : Specify a singularity image to use. The container image is searched for from the container registry directory environment variable in the swconf.yaml configuration.
Consider demo.swb, which is a batch script such as
or using a container image with a time of 42 hours
You can run the script as below but remember to export the container registry variable if you are using some custom singularity images.
Same as original slurm squeue, which show both running and queueing jobs, the swqueue shows running jobs and computational resource status.
Convert Tape ARchives to HDF5 files
- archive_checker: check how many files can be extracted from the input tar file.
- archive_checker_64k: check if any files within input tar files larger than 64 KB.
- h5compactor: convert input tar file into hdf5 file, all files within tar file should smaller than 64KB, using small files name as dataset names.
- h5compactor-sha1: convert input tar file into hdf5 file, all files within tar file should smaller than 64KB, using small files sha1 values as dataset names.
- h5shredder: convert input tar file into hdf5 file, no file size limitation, concatenate data and offsets into 4 seperate arrays for better randomized access.
Install dependent packages (on Ubuntu-20.04-LTS)
Install tar2h5 software
Uninstall tar2h5 software
Run tar2h5 software
Visualization with HDFCompass
Output File Format
- compactor output sample
- compactor-sha1 output sample
- shredder output sample