Bringing Old Giants Back to Life: The Ranger Cluster Story at Sol Plaatje University
The ranger cluster racks at Sol Plaatje University, once dormant, now reborn for modern research
What happens when you hand a determined team a room full of dormant, two-decade-old servers? At Sol Plaatje University (SPU), you get Ranger, a high-performance computing (HPC) cluster that proves innovation doesn’t always mean buying new.
Led by Gift Nemakonde and guided by Dr Silas Verkijika, this project is all about making the most of what you have. Instead of sending ageing Sun Blade hardware to the scrapyard, the SPU team rolled up their sleeves and got creative, breathing new life into forgotten machines with a modern software stack and a whole lot of grit.
Cluster By the Numbers
Nodes Online: 41 out of 48 possible (system can scale to 96)
CPU Power: 656 AMD Opteron cores (16 cores/node)
Memory: 1.3 TB RAM (32 GB/node)
Storage: 11 TB shared NFS storage
Headnode: Dell PowerEdge T340
Chassis: Sun Blade X6048 (2 racks, 8 chassis)
From Dusty to Dynamic
It wasn’t just a matter of plugging things in and flipping a switch. Each step was a battle against time, technical quirks, and the realities of legacy systems. The team rebuilt kernel drivers to restore dropped network connectivity, re-engineered the PXE/iPXE boot process to avoid RAM bottlenecks, and brought 41 of the 48 blades online, while work continues to bring the full system to stable production capacity. Instead of decompressing large OS images in memory, nodes now mount minimal images via NFS, slashing boot-time memory requirements and avoiding out-of-memory panics.
Modern Tech, Classic Machines
Ranger runs Rocky Linux 8, uses OpenHPC 2.x, and is managed with Warewulf 4 provisioning and Slurm as its scheduler. The cluster delivers all user software via centralised and stateless NFS shares, including advanced tools like GCC 12, OpenMPI 4.1.6, OpenBLAS, and Miniconda3 Python 3.12. Every detail is tuned for both reliability and efficiency.
Teamwork, Tenacity, and Ingenuity
Challenges weren’t just technical; they were logistical. Each blade required manual boot order configuration. Standardising Munge keys and enforcing NTP clock sync ensured node authentication and uptime. Moving to nolock NFS mounts eliminated frustrating mount failures and job script errors. Every problem was met with teamwork and creativity, turning setbacks into innovative solutions.
Performance That Defies the Calendar
Despite hardware dating back to 2008, Ranger punches far above its weight:
Monte Carlo Pi (1 billion iterations): 40x speedup across 41 nodes, maintaining 97.9% parallel efficiency
Matrix Multiplication (10,000 x 10,000): Completed reliably with no node failures
Network Latency: 38.6 microseconds (exceptional for 1 GbE)
Aggregate Write Throughput: 140 MB/s under parallel load
Ranger is now entering the stage of onboarding research projects across disciplines.
A Lesson in Making It Work
Ranger is more than just racks and cables. It’s proof that with enough skill and heart, yesterday’s technology can fuel today’s discoveries. At SPU, the Ranger cluster is a celebration of ingenuity, teamwork, and the belief that the best solutions often start with what you already have. As members of the HPC Ecosystems Project, we are proud to see these resources revitalised and delivering real impact for research and innovation at SPU.
#HPC #HighPerformanceComputing #Cluster #SPU #CHPC #OpenHPC #Linux #HPCecosystems #Innovation #Research
Continue Reading
