Dar used at IRISA

Who

IRISA (public research lab in computer science)

Context

Backup of self-managed scientific databases hosted at a local datacenter. The data was acquired over 15+ years using expensive equipment.

Start date

Year 2024

Backup/archive size

70To of objects saved with rclone
150Go of databases saved with dar

Media used

Why dar

It has supports for differential backups, binary diffs, catalogue isolation and is generally very tunable. The archive format is well documented and gives good confidence.

other remarks

We started with duplicity which worked off-the shelves, providing s3 storage, incremental backups, parity volumes and rotation. But it was not a perfect fit for our use case:

it only had support for incremental backups. We prefer differential backups for simplicity, robustness and the ability to set different expiration dates for daily vs. weekly backups. Also the incremental signature files were accumulated on the local disk (rather than consolidated), using around 15 times more space than necessary.
it did not suit our preferred S3 config. We wanted automatic object locking (to be ransomware-proof) and automatic expiration based on deletion time (NonCurrentVersionExpiration) for all buckets. Duplicity did not support restoring a backup from non-current objects, which would imply storing past backups longer than necessary. Also S3 objects were committed immediately after upload rather than at the end of the backup, which would uselessly consume space in case the backup is interrupted.
it did not allow tuning the compression parameters (we prefer zstd:1)
it did not allow tuning the librsync block size (2048 bytes only, which was smaller than necessary)

We did not find a libre alternative suiting all our needs and dar proved to be the lower-level and versatile tool that let us implement exactly what we wanted.

Source

Anthony Baire, March 2026

Back to page of references