Five clusters. How much work can we get done in 2 days?
On 16 and 17 April 2026 we will be getting together to work on our Single Board Computer (SBC) High Performance Clusters (HPC). At the moment we have five cluster that are being developed. These clusters have different purposes such as training, outreach and modelling. Come along to help us work on these clusters. If you have yet another cluster that can do with two days of hacking, contact us and perhaps it too can be included:
- CarpentriesOffline: RockPi cluster
- Jeremy’s DRIFT “Visible HPC” training cluster
- Sadie's CATS cluster
- CarpentriesOffline: Pixie RPi 4 cluster
- CarpentriesOffline RPi 5 cluster
CarpentriesOffline: RockPi Cluster
The RockPi Cluster is made up of six RockPi 4SE Single Board Computers. This cluster is intended to be used when access to an HPC for training purposes is difficult to obtain. This cluster needs the most work. Apart from 3D printing cases and a DIN rail for mounting, no attention has been given to the installation of the software. There are still quite a few unknowns with regards to the operating system and whether the cluster will be able to PXE boot or will need SD cards or SSDs to run from.
Jeremy's DRIFT "Visible HPC" training cluster
The "Visible HPC" mini-HPC portable training cluster is a 32+4-node Raspberry Pi 5 cluster developed to support the provision of introductory High Performance Computing training to a range of groups within the research community, from researchers to technical professionals to facilitators and engagement managers. The core motivation for the development of the cluster is that people learning about HPC today very rarely have the opportunity to see actual infrastructure. The days when people could tour on-site machine rooms at universities and see real HPC clusters in action are rapidly disappearing as HPC resources are increasingly housed off-site, in purpose-built data centres, or compute capacity is bought in from cloud providers. This can have significant cost and efficiency benefits but, for learners sitting in front of a terminal window on their laptop undertaking HPC training or using HPC resources, it’s very difficult to really understand what’s at the other end of the network connection!
This challenge started our “Visible HPC” activity that is being taken forward through a range of UKRI-funded projects. The cluster being brought to this workshop is our initial prototype portable training cluster that has been funded through the UKRI DRI-funded DRIFT project. We are developing a range of introductory training material alongside the cluster and have developed it to offer a number of properties that enable us to demonstrate key aspects of real data centres, just on a much smaller, and rather simplified, scale!
We hope that our “Visible HPC” cluster will help to make key aspects of now largely invisible HPC infrastructure visible again. We also hope it will inspire (and train) a new generation of potential HPC experts, engaging them in the same way that many of today’s HPC practitioners were engaged by visits to university machine rooms and the chance for hands-on experimentation with computer hardware.
CarpentriesOffline: Pixie RPi 4 cluster
When the miniHPC project started the intention was to use the RockPi SBCs. However, difficulties with ordering the hardware, at the time, led to using Raspberry Pi 4s which were already owned by Jannetta. The current status of the cluster is that it consists of one login node, five compute nodes and an access point. It reliably boots up, it uses PXE boot, POE, Slurm, munge, lmod and EESI. The most imminent things to be done is to create a backup image of the current SSD from which it is booting, to script adding PXE nodes, to script adding users and to stress test the cluster with learners using it for the Carpentries Intro to HPC lesson.
