Flexible Data Placement
Overview
Flexible Data Placement (FDP) is an NVMe mechanism that lets the host control where data is placed into different sets of physical SSD locations, so the host can maintain an optimal layout that reduces write amplification and garbage collection toward an SSD solution with write amplification factor (WAF) approaching one across all namespaces. The host steers data through reclaim unit handles into separate physical reclaim units, typically separating hot from cold data. FDP defines two isolation types — initially isolated and persistent isolated — where persistent isolation trades complexity for better write amplification on non-sequential light traffic.
FDP is positioned as the successful endpoint of a decade of data-placement attempts (open channel, streams, ZNS) because it is totally backwards compatible: it delivers benefits through configuration alone, without application rewrites. As of the 2025 OCP Global Summit it is ratified in NVMe Express, adopted into the OCP Datacenter NVMe SSD spec since revisions 2.5/2.6, shipping from Samsung and other vendors for 2-3 years, with Linux support across NVMe, the block layer, SPDK, and the XFS/EXT4/Btrfs file systems. Demonstrated gains include 70-80% WAF improvement through configuration changes, RocksDB drive WAF brought near 1.0, and Linux-kernel-build WAF cut from ~2.2 to ~1.0. Solidigm's multiple reclaim group FDP extends the design to up to 1,024 placement IDs, and ScaleFlux proposes per-reclaim-unit overprovision and resource isolation on top of it.
Sign in to read the full article.
Sign In