Understanding Data Protection: File‑Level vs Block‑Level Backup, Remote Replication, Snapshots, and Clone Technologies

Data Protection Overview

Data protection refers to backing up data at a specific point in time so that accidentally deleted data can be recovered. It can be divided into file‑level and block‑level protection.

File‑Level Backup

File‑level backup copies all files by invoking file‑system interfaces to another medium, reading data as files. The backup software only sees the file layer.

Files on the source medium may be fragmented, but after backup they can be stored contiguously, eliminating the need to back up metadata.

Block‑Level Backup

Block‑level backup copies every block of a block device regardless of file‑system logic, offering higher speed but also copying unused “zombie” sectors and preserving fragmentation.

Remote File Copy ( 远程文件复制 )

Remote file copy transfers files over the network to a disaster‑recovery site, typically using incremental synchronization (e.g., rsync).

Remote Volume Imaging

Based on block‑level backup, remote volume imaging copies block data to a remote site and can be synchronous or asynchronous.

Synchronous copy: The upper‑layer I/O is reported only after the data reaches the remote site.

Asynchronous copy: The write is acknowledged immediately; consistency is not guaranteed but response is faster.

Snapshot Technology

Snapshots capture the state of a volume at a specific moment. They rely on bitmap (BitMap) structures to record which blocks have been written since the snapshot.

Two main snapshot implementations are:

File‑System Based Snapshots

Use metadata such as inode tables, B‑trees, and bitmaps. Writing new data involves allocating free clusters, updating the file‑cluster mapping, and setting the bitmap.

Deletion only clears the mapping; the actual data remains until overwritten.

Physical‑Volume Based Snapshots

Record the original LBA mapping of a LUN; subsequent writes are redirected to new locations, with the original LBA table kept for reference.

Two snapshot write strategies are compared:

Copy‑on‑First‑Write (CoFW): On first write, the original block is copied to free space before being overwritten (one read, two writes).

Redirect‑on‑First‑Write (RoFW): The first write is redirected to a new location without copying (single write), but read performance may suffer due to bitmap lookups.

RoFW consumes more CPU resources, while CoFW consumes more I/O resources.

Clone Technology

A clone is a writable snapshot. A virtual clone shares unchanged blocks with the source, while a split clone creates an independent copy.

Backup Architecture

Backup components include the backup object, destination media, transport path, backup engine, and backup policy.

Backup destinations can be local disks, SAN, NAS, or virtual tape libraries, each with its own advantages and drawbacks.

Backup Paths

Local backup stays within the host; front‑end network backup traverses the client network; back‑end network backup uses SAN or dedicated links; LAN‑Free and Server‑Free backups avoid consuming front‑end resources.

Backup Strategies

Common strategies are full backup, differential backup, and incremental backup. Database backups usually rely on the database’s own tools.

For implementation details, see the original article.