Pages relating to the installation, maintenance and repair of the Ceph file/block distributed data system.
Historical notes of minor interest:
- DRBD. I first started running a distributed filesystem using DRBD. It’s fairly simple but it worked. However, since it was intended to run as a primary/failover system, it didn’t support triple redundancy. So I moved on to gluster.
- gluster. I liked gluster. It had relatively modest system requirements and allowed triple-redundancy mirroring. It could be touchy and somewhat esoteric to tune, but if all else failed, the backing store was essentially a traditional filesystem that could be accessed by brute force. But gluster is basically headed towards end-of-life. So I’ve migrated to ceph.
- Ceph is intended for large enterprise use. Its recommended minimal hardware requirements are considerable and only modern cheap hardware has made that truly viable for relatively small farms such as the mousetech.com R&D servers. Ceph comes with fairly good documentation (including lots of diagnostic and repair info) and some powerful tools. Its downsides are its relatively steep hardware requirements and the fact that you don’t actually know where your data is located. Not even which machine(s) it’s on!