Storage and Data Services
The storage model
AOS presents a Distributed Storage Fabric (DSF) across all nodes. Physical disks are pooled; on top of that you create storage containers - logical constructs that carry policy: replication factor, compression, deduplication, erasure coding. VM disks (vDisks) live in containers.
Services to know:
- Replication factor (RF2/RF3) - number of copies of every write.
- Erasure coding (EC-X) - parity-based space savings for colder data; it needs a cluster large enough to stripe across (four or more nodes) and increases rebuild impact.
- Inline compression - generally on by default and cheap.
- Deduplication - container-level, best for clone-heavy or repetitive data; not universal.
- ILM / tiering - moves cold extents to slower media and hot data to SSD.
- Curator - performs disk balancing and space reclamation in the background.
- Snapshots and clones - metadata operations, so they are near-instant and cheap until data diverges.
Traps
- Treating EC-X as a drop-in replacement for RF2 - it is a capacity optimisation for cold data, not a resilience model.
- Forgetting that a rebuild needs free space; a nearly full cluster may not be able to re-protect.
- Assuming dedup is always on. It is a per-container choice; compression is the one usually enabled by default.
- Confusing a storage pool (physical) with a storage container (policy).
Worked example
Four nodes, 4 TB HDD each = 16 TB raw. With RF2, usable is roughly half: about 8 TB, less metadata and rebuild reserve - plan on ~7 TB practical. Switching the same cluster to RF3 drops usable to roughly one third (~5 TB). Rule of thumb: usable ≈ raw ÷ RF, minus ~10% overhead.