Research

Tools

  • rdiff-backup
    • 1-way sync
    • Readable copy on backup server
    • keeps old versions, excellent restore capability
  • Unison
    • n-way merge (star topology)
  • Duplicity
    • Encrypted tar file remote backups
    • Incremental
    • Supports S3 etc.
  • Backupninja
    • high-level system backup agent with pluggable backup methods (duplicity, rdiff-backup) and sources (mysql, git, etc)
  • Git
    • n-way, incremental, requires explicit commits
    • ikiwiki provides a frontend for DVCS complexities

DHT

Idea: automatic replication of data to multiple nodes Use a tracker as redundancy mechanism.

Alternatives:

  • AFS
  • Chimera/Tapestry: http://current.cs.ucsb.edu/projects/chimera/

Inventory of Assets

Should be on a seperate page.

Documents

  • need n-way star: git (current) or unison
  • unison doesn’t provide versioning
  • git is labor-intensive Solved:
  • ikiwiki
    • supports star
    • auto-commit
    • wiki niceties (not backup related)

Plan

Idea: get all assets in one place, then use backup-ninja/rdiff-backup/duplicity on pool

Issues: * Existing systems: * n-way git hub ((sensichrome, silversurfer) > goober) * Idea: split out git repos into more granular repositories, throw away monolithic repo

Flow

On each system, a manifest lists everything to be backed up. A special group for sensitive data.

backup server: local backup => server/machinename

for servers:

  • Full-disk backups via rdiff-backup
    • Incremental
    • Simple to config
    • Needs a big disk (not a huge deal)
    • May take a long time to scan? (need to test)