Introducing GlusterFS

Sean Reifschneider

tummy.com, ltd.

Distributed File-Systems

• Wikipedia List of Distributed Filesystems
• Distributed:
  • Networked
  • NFS, CIFS/SMB, Andrew.
• Distributed Fault Tolerant
  • Replicated for availability and off-line access.
  • CODA, MS DFS, InterMezzo.
• Distributed Parallel
  • Hashed/distributed across multiple servers (RAID-0-like)
  • Lustre, PVFS.
• Distributed Parallel Fault Tolerant
  • Combine both of the above.
  • Gluster, Ceph, GPFS, Google FS, Hadoop DFS, future HAMMER...
• Distributed Peer-to-Peer
  • Loosely-coupled distribution.
  • Cleversafe

What is GlusterFS

• Easy to use.
• Parallel Fault-Tolerant File-System.
• Multiple nodes.
• Stackable modules that can be combined.
  • Striping, mirroring, replication.
  • BSDDB storage backend for small files.
• FUSE client-side.
• Simple file-system based back-end.
  • Easily recoverable outside of Gluster.

Future Functionality

• Live storage node add/remove.
• Hot software upgrade.
• Healing.
• Adaptive load-balancing of heavily used files.
• Active healing.
• Hash-based clustering (no name-space translator).

Example

• Demonstration time...
• /etc/glusterfs/glusterfs-{client,server}.vol
• /etc/init.d/glusterfsd restart
• glusterfs -f /etc/glusterfs/glusterfs-client.vol /mnt/glusterfs
• date >/mnt/glusterfs/foo
• date >/mnt/glusterfs/bar
• date >/mnt/glusterfs/baz
• find /data/gluster/{brick,brick-afr} -type f

References

• http://www.glusterfs.org/
• http://www.cleversafe.org/
• http://www.tummy.com/Community/Presentations/