JBOF Platform.

Just-a-Bunch-of-Flash, engineered properly.

A pure-flash storage chassis with no in-path CPU. PCIe Gen4 fabric fans out to 24 NVMe drives, delivering deterministic latency and dense capacity for AI feature stores and training datasets.

Because there's no in-path CPU, the JBOF Platform delivers PCIe-native latency — a few microseconds from host issue to drive completion. Software-defined storage (SDS) or Ceph/OpenEBS clusters can consume the drives directly.

Detailed description.

A JBOF is the simplest form of shared storage — a chassis full of drives, connected via a fabric, exposed as raw NVMe namespaces. No RAID controllers, no in-path CPU, no proprietary software layer. The JBOF's job is to be transparent.

The Datafabrix JBOF Platform uses our Gen4 Fabric Backplane to switch PCIe lanes from two host cables to 24 downstream NVMe bays. From the host's perspective, each drive is a directly-attached NVMe device — with all the latency and IOPS characteristics you'd expect from a local PCIe slot.

Because the platform is fabric-transparent, it drops into existing Ceph, OpenEBS, MinIO, or in-house SDS stacks without changes. You get scalable capacity behind PCIe Gen4 latency — and the operational simplicity of raw NVMe.

Applications

• Ceph pools — Dense NVMe capacity for Ceph OSDs on shared fabric.
• OpenEBS cStor — SPDK-accelerated block storage on top of NVMe.
• Object storage — MinIO / SwiftStack backends with predictable latency.
• AI checkpoints — High-write-endurance NVMe for periodic training snapshots.
• Warm data caches — Fast-tier caching in front of slower object stores.