Login Node Usage

HPC login nodes serve as an access point for users wishing to run jobs on the ITS-RS HPC system. When you log in, you will be placed in a round-robin fashion on one of the available login nodes.

Please do not run your production research computations on the login nodes. Login nodes are limited in number and shared by all users. Compute nodes are dedicated to running job processes. Please run computationally intensive processes there instead, preferably by submitting job scripts or, if appropriate for your work, by using a single qlogin session. HPC staff will terminate any long running processes found running on a login nodes, with the exception of localized job control scripts that are sometimes needed.

Login nodes are intended for basic tasks such as uploading data, managing files, compiling software, editing scripts, and checking on or managing your jobs. Small-scale interactive work may also be acceptable if your resource requirements are minimal. Remember that when you are using a login node, there are generally many other people also using it at the same time. Using too many resources on a login node may negatively affect the ability of other people to complete their work.

To help prevent this, login nodes have a per-user CPU quota which limits total simultaneous usage by all processes to 100% of a single logical CPU. Login nodes also have a memory limit of 16G per user for all processes, so please be aware that if you run processes on a login node that push the total beyond a certain amount, the limiter will begin killing the largest processes until the total again satisfies the limit.

Finally, we reserve the right to terminate any process that we feel is clearly interfering with the ability of others to complete work, regardless of technical measures or its resource usage.

By default, Java uses a maximum heap size based on a large fraction of the physical memory it detects. The default is always larger than the above mentioned limits on login nodes, so need to explicitly specify an appropriate maximum heap size for your application. Try the following as a starting point:

java -Xmx1g ...

DUO

Duo two-factor Authentication is required for cluster access. Make sure you have configured Duo in order to log in. More info here: Two-factor authentication

cron jobs

You may want to use cron jobs to automate reoccurring aspects of a workflow, such as checking for new data or submitting cluster jobs to process it. Argon's login nodes share a cron directory so you can manage all your cron jobs in one location, and a single login node will run them. You can use cron syntax and features supported by the "cronie" implementation of cron present in CentOS 7.

SGE in cron

Note that cron's default environment is unlike normal shell sessions and lacks some typical features and configuration. In particular, if you use any SGE commands in your cron jobs or in scripts you call, you first need to configure SGE's environment variables by sourcing whichever sge.* script in /etc/profile.d is appropriate for your shell interpreter. Versions are available for Bourne and C family shells:

/etc/profile.d/sge.sh
/etc/profile.d/sge.csh