Speaker
Description
The continuous growth in data volumes and diversification of access patterns in high-energy physics (HEP) are gaining attention in storage systems that offer both extreme performance and ease of use. To explore the potential of modern flash technologies for scientific workloads, we conducted a comprehensive benchmarking campaign for a PureStorage all-flash appliance, focusing on its suitability for few key physics data processing scenarios.
The study evaluates the system’s performance using representative HEP I/O patterns, including sequential and random access, metadata-intensive workloads with small files, and mixed read/write parallelism, as are typical for user analysis and detector reconstruction tasks. Using standard benchmarking tools and custom workflow emulations, we evaluated throughput, latency, and scalability under varying degrees of parallelism and dataset complexity.
Beyond raw performance, we evaluated multi-protocol access (NFS, SMB, S3) and executed a typical IO500 HPC benchmark on the appliance.
This evaluation delivers insights on how all-flash storage can complement existing HEP infrastructures for enterprises, identifying both performance benefits and architectural trade-offs relevant to future data centre designs and hybrid storage systems within the physics computing community. This becomes particularly interesting in a multi-tier configuration where EOS uses a disk-based, erasure-coded storage tier in combination with an all-flash storage tier.