Skip to content

Latest commit

 

History

History
79 lines (66 loc) · 3.13 KB

SLURM_Instructions.adoc

File metadata and controls

79 lines (66 loc) · 3.13 KB

Running GGCMI-Weather on SLURM-based systems

How it works

Configuration

Configuration is handled in the config.py script included in the application directory. The configuration consists of the following:

Top Level Variables

start_date

a python datetime object used to indicate the starting date of the provided NetCDF files.
station_vector

the vector file used to bind a latitude and longitude to the weather staiton.
station_id_field

the field within the vector file used to indicate the file name of the weather station.
output_dir

the directory into which the weather files will be generated.

Mapping variables

Each object underneath the mapping block consists of the following:

file

The NetCDF file to extract data from
netcdfVar

The variable to extract from the NetCDF file
dssatVar

The variable the netcdfVar represents in the DSSAT weather file.
conversion

A lambda used to convert the unit inside the NetCDF file to the appropriate DSSAT unit.
Note
 The order in which the mapping is defined is the column order in the DSSAT Weather file.

Start the server

  1. Edit the slurm-runserver.sh file.

    The only pieces to modify should be --cpus-per-task or --time. --cpus-per-task should be set between 10-15. --time should be set over the expected length of time to execute.

  2. Submit the slurm-runserver.sh job.

    > sbatch slrum-runserver.sh
Submitted batch job 1234567

  3. Wait for the server to actually start. This can be checked using the squeuemine from the ufrc module. When ST is R then the server is running.

    > ml ufrc
> squeuemine
             JOBID    PARTITION              NAME       USER ST       TIME  NODES NODELIST(REASON)
             1234567  hpg2-comp         GGCMI-SRV cvillalobo PD       0:00      1 (None)
> squeuemine
             JOBID    PARTITION              NAME       USER ST       TIME  NODES NODELIST(REASON)
             1234567  hpg1-comp         GGCMI-SRV cvillalobo  R       0:45      1 c6a-s29

  4. Retrieve the IP address from the sbatch log file: (Default is: slurm-<jobid>.out)

    > cat slurm-1234567.out
172.16.199.29 : 7781
Python 3.7.6
Server data initializing.
> Number of valid points: 65797

Start the clients

  1. Edit the slurm-runclients.sh

    To configure the number of nodes/clients you want generating files set both the #SBATCH --nodes and client_num to the same value. You need to configure the ip_address to the same as from the last step in Start the server.

    Note
    		 You can always join more nodes to the server later. It takes less resources to start a small number of clients and will more likely get your job allocated. Just submit the slurm-runclients.sh again.

  2. Submit the slurm-runclients.sh job.

    > sbatch slurm-runclients.sh
Submitted batch job 12345679

As soon as the server initialized all the data, the clients will automatically start processing the weather files. When the server runs out of data to serve, it will automatically shut down, as will all the clients.