What is JBOD?: rawcompute.in Glossary

JBOD stands for “Just a Bunch of Disks” and refers to a storage configuration where multiple drives are presented to the host operating system individually, without any RAID controller grouping them into arrays. Each drive operates as a standalone block device. This approach is used when the application layer handles data redundancy (e.g., HDFS, Ceph, MinIO) or when drives are used as ephemeral scratch space where data loss is acceptable.

In GPU server configurations, JBOD is common for NVMe drives used as local scratch storage during training. Training datasets are typically stored on a shared distributed file system, and local NVMe drives serve as a high-speed cache. If a drive fails, the node simply re-fetches data from the shared storage. JBOD eliminates the capacity overhead of RAID parity and avoids the performance penalty of parity calculations.

Why it matters when buying hardware

Use JBOD when your application provides its own data redundancy or when drives are used for temporary/cache storage. Many Hadoop, Ceph, and MinIO deployments specifically require JBOD because they implement replication at the software layer. Adding RAID underneath wastes capacity and can hurt performance. When ordering servers from rawcompute.in, specify whether you need a RAID controller or an HBA (Host Bus Adapter) in JBOD/passthrough mode. Some RAID controllers can be flashed to IT mode for JBOD use.