Glossary

ARC (Adaptive Replacement Cache) — ZFS's read cache in RAM: recent and frequent data, served from memory. Grows into free RAM, shrinks under pressure. Chapter 2.

ashift — the pool's record of the disks' sector size, set at creation (12 = 4K sectors) and unchangeable after. Wrong guesses cost performance permanently. Chapter 4.

Bookmark — a marker standing in for a snapshot: keeps the position an incremental send needs, without pinning the snapshot's data. Written with # where a snapshot uses @. Chapter 7.

Checksum — a fingerprint of a block, stored away from the block, verified on every read. The heart of ZFS. Chapters 1 and 2.

Clone — a writable dataset created from a snapshot; shares its blocks until they diverge. Chapter 6.

Copy-on-write — writes go to fresh space and pointers flip atomically; old versions remain until freed. Crash-safe by construction, and the reason snapshots are cheap. Chapter 2.

Dataset — a named, separately managed slice of a pool, with its own properties, snapshots and quota. Looks like a folder, acts like a filesystem. Chapter 5.

Dedup (deduplication) — storing identical blocks once, pool-wide. RAM- hungry; this guide advises leaving it off. Chapter 5.

L2ARC — a second-level read cache on a fast device (usually an SSD), fed by the ARC. A "cache vdev". Chapter 2.

Mirror — a vdev whose disks each hold everything; survives all but one of them failing. Chapter 2.

Pool (zpool) — all the storage ZFS manages together, built from vdevs, divided into datasets, referred to by a name like tank. Chapter 2.

Property — a named setting on a pool or dataset (compression, quota, ...). Inherited from parents unless set locally. Chapter 5.

RAIDZ1 / 2 / 3 — parity vdevs: data plus one, two or three disks' worth of parity, surviving that many disk failures. Chapter 2.

Recordsize — the largest block a dataset writes in one piece; a property to match to the workload. Chapter 5.

Resilver — rebuilding a replaced disk from the surviving copies: only live data is copied. Chapter 8.

Rollback — rewinding a dataset to a snapshot, discarding everything written after it. Chapter 6.

Scrub — a full read-and-verify of the pool, repairing from redundancy; run on a schedule. Chapter 8.

Send / receive — writing a snapshot to a stream, and rebuilding a dataset from such a stream: ZFS's replication. Chapter 7.

SLOG — a separate log device (a "log vdev") absorbing synchronous writes. An optimisation most homes don't need. Chapter 2.

Snapshot — a read-only, point-in- time view of a dataset, kept by copy-on-write at near-zero cost until the data diverges from it. Chapter 6.

Vdev (virtual device) — a group of disks in one shape — mirror, RAIDZ, single — joined to a pool. Redundancy lives at this level. Chapter 2.

ZED — the ZFS Event Daemon: watches pool events and can run scripts or send mail when the pool needs a human. Chapter 9.

Zvol — a dataset presented as a raw block device, for VM disks and iSCSI. Named in chapter 2; this guide stays with file datasets.