Quick start¶

A first job on Hydra takes five steps. You make a working directory, write a job file, submit it with qsub, check on it with qstat, and read the output file at the end. You need an account and to be logged in. The example job needs no input data and runs for 30 seconds.

Replace USERNAME with your Hydra username throughout.

1. Log in and create a working directory¶

Jobs run from a directory under /scratch. The home directory has a small quota and is not for job input and output.

Your directory on /scratch is under a group directory such as /scratch/genomics, /scratch/sao or /scratch/odi. Your welcome email tells you which one is yours. The examples on this site use /scratch/genomics/USERNAME. Substitute your own path.

ssh [email protected]
mkdir -p /scratch/genomics/USERNAME/quickstart
cd /scratch/genomics/USERNAME/quickstart

2. Write a job file¶

A job file is a shell script with extra lines at the top. Lines beginning with #$ are options to the scheduler, and the rest is what the job runs. Create hello.job with a text editor (nano hello.job if you have no preference):

hello.job

#!/bin/sh
#$ -S /bin/sh
#$ -l mres=2G,h_data=2G,h_vmem=2G
#$ -cwd
#$ -j y
#$ -N hello
#$ -o hello.log
#
echo + `date` job $JOB_NAME started in $QUEUE with jobID=$JOB_ID on $HOSTNAME
sleep 30
echo = `date` job $JOB_NAME done

Option	Effect
`-S /bin/sh`	Shell used to run the script
`-l mres=2G,h_data=2G,h_vmem=2G`	Reserve 2 GB of memory
`-cwd`	Run in the directory the job was submitted from
`-j y`	Write error output to the same file as standard output
`-N hello`	Job name
`-o hello.log`	Output file

There is no -q line, so the job goes to the default queue, sThC.q, which allows up to 7 hours of CPU time and is right for most first jobs. Job script reference lists every option, and the QSub Generator writes a job file for you from a web form.

3. Submit the job¶

$ qsub hello.job
Your job 15497588 ("hello") has been submitted

The number is the job ID. qstat, qdel and qacct refer to jobs by this ID, and so will we if you write to us about a job.

4. Check the job¶

$ qstat
job-ID     prior   name       user         state submit/start at     queue                          jclass                         slots ja-task-ID
------------------------------------------------------------------------------------------------------------------------------------------------
  15497588 0.50500 hello      USERNAME     r     09/30/2026 18:48:14 [email protected]                                    1

State qw means queued and waiting, r means running. A job with an error in its job file usually fails within seconds, so check qstat during the first minute after submitting. When the job has finished, qstat prints nothing.

You can log out with exit while a job is running. The scheduler owns the job, so it keeps running after you log out.

5. Read the output¶

$ cat hello.log
+ Wed Sep 30 18:48:14 EDT 2026 job hello started in sThC.q with jobID=15497588 on compute-93-02
= Wed Sep 30 18:48:44 EDT 2026 job hello done

/scratch is not permanent storage

The scrubber deletes files older than 180 days. Move results off the cluster when an analysis is complete. See Storage.