Ranking
Note: These Benchmarks are carried out by different companies in different environments, so results may differ on other systems. I did find similar results in other benchmarks and selected these ones because they closely resemble the HPC we use at Tubklima.
Publication note: this public summary avoids reporting precise preliminary benchmark values from active work.
Lustre: In external benchmark reports, Lustre commonly appears in the highest-throughput group for parallel read and write workloads in distributed storage setups.
BeeGFS: BeeGFS is frequently reported as a strong performer for mixed throughput and metadata-oriented HPC workloads, depending on topology and tuning.
GlusterFS: GlusterFS is often selected for operational flexibility and simpler deployment models, with throughput profiles that vary strongly by workload and configuration.
Ceph: Ceph provides a robust unified storage stack and is commonly favored for feature breadth and resilience, with performance depending on hardware, networking, and tuning strategy.
Practical takeaway
Platform choice should be based on workload profile (metadata-heavy vs throughput-heavy), operational constraints, and reliability goals rather than on isolated headline numbers.
See also: BeeGFS course overview, MOC Computing DevOps and Software