DAS (Direct-Attached Storage)
Direct-Attached Storage (DAS) is a storage configuration where storage devices are directly connected to a single server via PCIe, SAS, or SATA — not shared over a network. Local NVMe SSDs in a server are a form of DAS. While DAS delivers maximum single-node performance, it prevents storage sharing, limits HA options, and ties capacity expansion to server upgrades.
DAS vs Disaggregated NVMe-oF
| Attribute | DAS (Local NVMe) | Disaggregated NVMe-oF |
|---|---|---|
| Latency | 5–20µs | 25–40µs (NVMe/TCP) |
| Shared access | No | Yes — multiple hosts |
| HA / replication | None (single node) | Multi-path, N-way replication |
| Scale storage independently | No | Yes |
| Kubernetes PVC support | Limited (local PV) | Full CSI PVC support |
| Data survives host failure | No | Yes |
When DAS Makes Sense
DAS (local NVMe) remains the right choice when absolute minimum latency is required and high availability is handled at the application layer — for example, Cassandra or CockroachDB nodes that replicate data at the database level. For most Kubernetes workloads, however, disaggregated NVMe-oF delivers a better trade-off: 25–40µs latency with HA, PVC mobility, and independent scaling.