The command jobinfo is able to give you relevant information about the performance of your job.
For example you ran a job with the job ID 12345. Then use the following command in order to see this information:
jobinfo 12345
For more detailed information about the performance of your job, see using_jobinfo
In order to test small jobs on the cluster, you can log in on the interactive node of the Hábrók cluster. By typing
ssh username@interactive1.hb.hpc.rug.nl
you will be prompted to enter your password. From here you can execute small programs by simple executing them as you would do in a normal terminal. Since the interactive node is shared among other users, a small job should take up to ±30 minutes and should not be using all available resources (e.g. CPUs, memory).
A few Python modules are already installed on the Hábrók cluster, however if you want to use a module that is not installed yet, you can install it for your own user account. The Hábrók cluster automatically will detect that you have that package installed in your personal directory (/home/username/) where your username is something like “p123456”. Installing a module is described on this page, but here is an short answer:
module load Python/3.5.1-foss-2016a pip install --user packageName
Make sure that the P from Python is capitalized.
In order to install a package in R, you have to start R in the cluster. This is done by loading the module R, starting R (by typing “R” in the terminal) and install the package as you normally install a package in R. After the installation you can quit R. The Hábrók cluster now will detect that you have this package installed when it is running a job which makes use of that particular package.
Depending of the amount of resources that are requested for a job, it can take a while for a job to get started. Therefore, it might be useful to get a notification of when the job is started. It also is very helpful to know when a job is completed if you need the results of the job for further research. With the information of the begin and completion time of a job, a better estimation can be made for running another similar job.
Each CPU on a 128GB node has about ~5GB memory (128GB/24cores), therefore multiplying this value with the number of cores that are requested is a good way to start. If this is not enough, your job will crash with an out-of-memory error; you then have to rerun it and increase the memory request (for instance, try setting the limit twice as high). After a job has run, you can see in the summary at the bottom of the output file (or by using jobinfo <job id>) how much memory it actually used.
The login node has to be used by all users of the cluster for connecting to the cluster, copying data, submitting jobs, et cetera. Therefore, the login node has some user limits on the CPU and memory usage to prevent one user from taking all these resources. If you are running something, you may exceed this limit and your application will be killed. If this happens you could try to run this process on the interactive node, which has more resources and less stringent limits, or you can submit this task as a job.
Yes, this is possible. It can be done by simply providing the parameters like sbatch script.sh parameter1 parameter2. Look here for an example with Python.
Please make sure that you use lowercase usernames. For instance, use p123456 or s123456 instead of P123456 or S123456. Also, make sure that you can still log in to other university services (e.g. your email); if those still do work, contact us at hpc@rug.nl.
When the error message contains something like:
/tmp/cceBXn0u.s: Assembler messages: /tmp/cceBXn0u.s:1677: Error: no such instruction: `shlx %rax,%rdx,%rax'
You can try installing this package manually on the login node in order to solve this problem (e.g. for R or Python). Otherwise, you need to load the right binutils module (e.g. module load binutils/2.25) before installing this package or module.
After you have submitted your job, your job is put into a queue. To check the status of your job, you can type:
squeue -u $USER
or:
jobinfo <job id>
When your requested resources are available to you, your job is started.
Some more information about your position in the queue and the estimated start time of your job can be found here.
The Hábrók cluster uses a priority queue. When you submit a lot of jobs, your priority decreases. This way, other users also get their share of the cluster. How this prioritization is done, can be found here.
There are some limits that prevents the cluster from being filled with a lot of long jobs, which would cause long waiting times for all other jobs. Jobs requesting more than three days of wall clock time in the “regular” partition are considered to be long jobs. So, if your job has this state, it will have to wait for some other long job(s) to finish.