Running Simulations

In this section:

Basic Workflow
First Job: Simple Example
Monitoring Jobs
Queue System (Partitions)
File Systems
Basic Command Summary

This section presents the fundamental concepts for processing jobs on GridUnesp, from environment preparation to execution monitoring.

Important

The server access.grid.unesp.br is exclusively for access and preparation.

NEVER run simulations directly on access. Always use the queue system (SLURM) to submit jobs to the processing nodes.

Basic Workflow 

The typical workflow on GridUnesp follows these steps:

Access the access server (ssh)
Prepare your files and scripts in /home/
Submit the job with sbatch
Monitor execution with squeue
Retrieve results after completion

First Job: Simple Example 

Let’s create a test job that simply displays the date and waits 60 seconds.

Step 1: Create the submission script

first_job.sh

#!/bin/bash
#SBATCH -t 5:00        # Maximum time: 5 minutes

date
sleep 60

Step 2: Submit the job

sbatch first_job.sh

Step 3: Check the status

squeue -u $USER

Step 4: View the result

cat slurm-*.out

Note

The output file follows the pattern slurm-JOBID.out.

Monitoring Jobs 

squeue command 

# View all jobs
squeue -a

# View only your jobs
squeue -u $USER

# View jobs in detailed format
squeue -u $USER -o "%.18i %.9P %.8j %.8u %.2t %.10M %.6D %R"

Job states:

PD (PENDING): Waiting for resources
R (RUNNING): Running
CG (COMPLETING): Finishing
F (FAILED): Failed
TO (TIMEOUT): Exceeded time limit

Example output:

JOBID    PARTITION        NAME   USER  ST      TIME  NODES  NODELIST
2848879      short  vulcano.sh  spock   R      0:09      1   node011
2841620       long   crucio.sh  vader   R  10:02:40      1   node018

scontrol command 

To get full details of a job:

scontrol show job JOBID

Example partial output:

JobId=2848879 JobName=vulcano.sh
  UserId=spock(11992) GroupId=enterprise(11006)
  JobState=RUNNING
  RunTime=00:00:18 TimeLimit=00:05:00
  NodeList=node011
  Command=/home/spock/vulcano.sh

scancel command 

To cancel a job:

scancel JOBID

To cancel all your jobs:

scancel -u $USER

Queue System (Partitions)

GridUnesp organizes jobs into partitions based on maximum execution time:

Available partitions
Partition	Limit	Recommended use
short	24 hours	Tests, quick jobs (default)
medium	7 days	Medium-duration simulations
long	30 days	Long simulations
gpu	24 hours	Exclusively for the GPU server

Important

If not specified, the job goes to the short partition with a 24h limit
When the limit is reached, the job is automatically cancelled (TIMEOUT)
Always specify a realistic time (shorter jobs have priority)

Specifying time:

#SBATCH -t 30:00                 # 30 minutes
#SBATCH -t 12:00:00              # 12 hours
#SBATCH -t 2-12:00:00            # 2 days and 12 hours
#SBATCH -t 30-00:00:00           # 30 days

File Systems 

During execution, jobs use different storage areas:

Mount points
Location	Capacity	Use
`/home/`	120 TB	Job preparation (accessible from access server)
`/tmp/`	~180 GB/node	Single-node jobs (fast, temporary)
`/store/`	7 TB	Multi-node jobs and large files

Tip

See Complete Storage Guide for details on how to optimize the use of each area.

Basic Command Summary 

Essential commands
Command	Description
`sbatch script.sh`	Submit job
`squeue -u $USER`	View your jobs
`scancel JOBID`	Cancel job
`scontrol show job JOBID`	Job details
`sacct -j JOBID`	Job history (after completion)