You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
Dennis Schwerdel f9ab8e8cc7 Version 0.1.0 5 years ago
deb Version 0.1.0 5 years ago
docs Better control over what is checked in `check` subcommand 5 years ago
src Fixes and progress bars 5 years ago
.gitignore First packaging 5 years ago
Cargo.lock Version 0.1.0 5 years ago
Cargo.toml Fixes and progress bars 5 years ago
LICENSE.md First packaging 5 years ago
README.md Some fixes 5 years ago
TODO.md Dirty flag on error 5 years ago
gpl-3.txt First packaging 5 years ago
test.sh Updated comparison 5 years ago

README.md

zVault Backup Solution

zVault is a highly efficient deduplicating backup solution that supports client-side encryption, compression and remote storage of backup data.

Main Features

Space efficient storage

Each file is split into a number of chunks. Content-defined chunking and chunk fingerprints make sure that each chunk is only stored once. The chunking algorithm is designed so that small changes to a file only change a few chunks and leave most chunks unchanged. Multiple backups of the same data set will only take up the space of one copy.

The deduplication in zVault is able to reuse existing data no matter whether a file is modified, stored again under a different name, renamed or moved to different folder.

That makes it possible to store daily backups without much overhead as backups with only small changes do not take up much space.

Also multiple machines can share the same remote backup location and reuse the data of each others for deduplication.

Performance

High backup speed is a major design goal of zVault. Therefore is uses different techniques to reach extremely fast backup speeds.

All used algorithms are hand-selected and optimized for speed.

Unmodified files are detected by comparing them to the last backup which makes it possible to skip most of the files in regular usage.

A blazingly fast memory-mapped hash table tracks the fingerprints of all known chunks so that chunks that are already in the repository can be skipped quickly.

In a general use case with a Linux system and a home folder of 50 GiB, backup runs usually take between 1 and 2 minutes.

Independent backups

All backups share common data in form of chunks but are independent on a higher level. Backups can be deleted and chunks that are not used by any backup can be removed.

Other backup solutions use differential backups organized in chains. This makes those backups dependent on previous backups in the chain, so that those backups can not be deleted. Also, restoring chained backups is much less efficient.

Data encryption

The backup data can be protected by modern and fast encryption methods on the client before storing it remotely.

Compression

The backup data can be compressed to save even more space than by deduplication alone. Users can choose between zlib (medium speed and compression), lz4 (very fast, lower compression), brotli (medium speed, good compression), and lzma (quite slow but amazing compression).

Remote backup storage

zVault supports off-site backups via mounted filesystems. Backups can be stored on any remote storage that can be mounted as a filesystem:

  • NFS
  • SMB / Windows shares
  • SSH (via sshfs)
  • FTP (via curlftpfs)
  • Google Drive (via rclone)
  • Amazon S3 (via rclone)
  • Openstack Swift / Rackspace cloud files / Memset Memstore (via rclone)
  • Dropbox (via rclone)
  • Google Cloud Storage (via rclone)
  • Amazon Drive (via rclone)
  • Microsoft OneDrive (via rclone)
  • Hubic (via rclone)
  • Backblaze B2 (via rclone)
  • Yandex Disk (via rclone)
  • ... (potentially many more)

Backup verification

For long-term storage of backups it is important to check backups regularly. zVault offers a simple way to verify the integrity of backups.

Mount backups as filesystems

Backups can be mounted as a user-space filesystem to investigate and restore their contents. Once mounted, graphical programs like file managers can be used to work on the backup data and find the needed files.

Example usage

As an example, I am going to backup my projects folder. To do that, I am initializing an encrypted zVault repository, storing the data on a remote filesystem which has been mounted on /mnt/backup.

#$> zvault init :: --encrypt --remote /mnt/backup
public: 2bea1d15...
secret: 3698a88c...

Bundle size: 25.0 MiB
Chunker: fastcdc/16
Compression: brotli/3
Encryption: 2bea1d15...
Hash method: blake2

The repository has been created and zVault has generated as new key pair for me. I should now store this key pair in a safe location before I continue.

Now I can backup my home directory to the repository.

#$> zvault backup /home/dswd/projects ::projects1
info: No reference backup found, doing a full scan instead
Date: Thu,  6 Apr 2017 20:33:20 +0200
Source: dswd-desktop:/home/dswd/projects
Duration: 0:00:26.2
Entries: 14618 files, 6044 dirs
Total backup size: 1.4 GiB
Modified data size: 1.4 GiB
Deduplicated size: 1.2 GiB, 14.9% saved
Compressed size: 0.5 GiB in 23 bundles, 54.7% saved
Chunk count: 95151, avg size: 12.8 KiB

The backup run took about 26 seconds and by looking at the data, I see that deduplication saved about 15% and compression again saved over 50% so that in the end my backup only uses 0.5 GiB out of 1.4 GiB.

After some work, I create another backup.

#$> zvault backup /home/dswd/projects ::projects2
info: Using backup projects1 as reference
Date: Thu,  6 Apr 2017 20:46:19 +0200
Source: dswd-desktop:/home/dswd/projects
Duration: 0:00:00.7
Entries: 14626 files, 6046 dirs
Total backup size: 1.4 GiB
Modified data size: 27.2 MiB
Deduplicated size: 17.2 MiB, 36.9% saved
Compressed size: 6.1 MiB in 2 bundles, 64.4% saved
Chunk count: 995, avg size: 17.7 KiB

This time, the backup run took less than a second as zVault skipped most of the folder because it was unchanged. The backup only stored 6.1 MiB of data. This shows the true potential of deduplication.

Semantic Versioning

zVault sticks to the semantic versioning scheme. In its current pre-1.0 stage this has the following implications:

  • Even now the repository format is considered pretty stable. All future versions will be able to read the current repository format. Maybe conversions might be necessary but the backups should always be forward-compatible.
  • The CLI might see breaking changes but at least it is guaranteed that calls that are currently non-destructive will not become destructive in the future. Running todays commands on a future version will not cause any harm.