1-0. Job scheduling basics

Job scheduling and batch jobs

The parallel servers of the JAIST use the job scheduling system.

The  system automatically executes the computational jobs as soon as the resources are available. The computation are registered as batch jobs collecting the executable files, data, etc. required for the calculation and considered as a unit to the job scheduling system.

Each server is split into a Login node and computation nodes user packs the executable file and data for calculation at the login node and submits it to the job scheduler.

Job scheduling basics

Basic usage procedure

1. Making the job script
Executable files, data, setting files etc., which are necessary for execution of the job are packed and listed in a routine job script

2. Job submission
Submit the job script to the job scheduler. There is no limit to the number of jobs. Any number of jobs with different input data can be submitted.

3. Running the job
A computation job is automatically executed when the resources (CPU, Memory, Node, etc.) necessary for execution are available

Rules of running the job

Jobs that have been submitted to the scheduler are executed according to the following rules:

・Tasks that can be executed with less resources are prioritized

・First come first serve for tasks input in the same queue

・First come first serve for tasks submitted in the same queue

・Tasks that have not been executed for a certain period (due to insufficient resources etc.) are executed as top priority.
In this case, other jobs may not be started temporarily in order to secure resources for executing the highest priority job.