Skip to content

Mathematics & Computer Science

About the hpc.tue.nl-cluster

Onboarding documentation: a mandatory read for all users

Introduction

The hpc.tue.nl-cluster is a cluster of compute resources (CPU/GPU) that is made available for research staff. The facility originates from a collectively expressed need by M&CS research in 2019 for low-barrier (close to home but serviced) compute resources that could facilitate most of the computations. The main goal was to enable an experimental environment to test jobs[BB3] where one would not have to worry about unnecessarily burning rented compute hours from external resources. When using the system, be aware of the impact your type of job has on the experience of others that you share the hardware with: this document helps you with this.

Disclaimer

The cluster is essentially a fixed and shared set of hardware by all of the research department of M&CS. Policy states that all hardware must remain available for everybody at all times: this means isolation of machines for individual cases is not possible. Also take into account, you can expect that there are waiting times (of unexpected length), improper use of resources by others and that in the end, the cluster might not be a suitable place for your type of job. Know that we do have alternatives (please contact your Research IT consultant for the options).

Relevant technical aspects

All specifications of the machines in the cluster are stated here. There is a quota of 1 TB on every user's home directory. Because there is roughly 95TB in total, there is likely no room for everyone to fill up their quota completely. Do not use the cluster as a place to store data . Every compute node does have slightly over 2 TB of scratch storage available under /local, which does not fall under the quota (but is deleted once a job finishes). The data residing on the HPC cluster has no back-up!

Documentation

You can browse the HPC wiki and read about how to login, transfer files to and from the cluster, define and submit compute jobs, track their progress, work with specific tools/software, et cetera.

Community

We have a mailinglist and a Teams environment for all communication concerning this cluster. Please subscribe if you want to think along about cluster development/expansion; know about technical changes or outages; be informed about workshops and hand-on sessions, etcetera. Next to that, feel free to reach out to your co-users when you have doubts about their way of using the cluster.

Specialized support

Support on your code and/or scripts is available! Contact HPC lab for a consultant.

User guidelines

GENERAL PRACTICES

  • Please consider at any time the impact of the job you’re about to run (given (expected) wall time, allocation of resources, etc.). Be especially considerate of running jobs that need a disproportional amount of cores w.r.t. what has been allocated (i.e., if your allocation of resources in your script does not reflect what you actually use)): this prevents others from using the unused resources.
  • Do not compile code on the log-in node.
  • Practical suggestions, tips, and practices on how to use the cluster are collected and shared on this website. For example, are you using Tensorflow on the GPUs? Read here for best practices.

In case you need compute resources for:

  • …education,
  • …benchmarking,
  • …working with a virtual machine,
  • …compute a job that needs a lot of RAM,
  • …jobs that take more than 7 days on the cluster,
  • …development,
  • …jobs that cannot run multi-threaded,

The cluster might not be the best fit for your purpose. Please contact hpcsupport@tue.nl to discuss other options to consider with comparable functionalities and reasonable border conditions.

Rules

(TECHNICALLY) ENFORCED RULES ON CLUSTER

  • Required to specify the number of cores/RAM explicitly for GPU queue
  • Fixing an equal distribution for every user w.r.t. number of jobs scheduled
  • Implementing a priority system (FAIR system). As this is WIP, a cap (=3) on the number of nodes one can claim will be enforced
  • A limit on wall time (being 7 days)