5 comments

[ 5.0 ms ] story [ 25.2 ms ] thread
How does ceph compare to glusterfs, POHMELFS, sector/sphere etc ?

Is anybody here using this ?

First you need to distinguish between the ones that just give you some kind of blob api (HDFS, mogilefs, whatever sector/sphere's FS component is), and mostly-posix-compliant ones (glusterfs, ceph, etc).

http://pl.atyp.us/wordpress/?p=2662 gives some details on the 2nd category, so that's probably more useful in a discussion about ceph.

Sage Weil, the creator of Ceph, recently gave a talk at linux conf au. The slides [1] mostly stand on their own. The first two slides address your first question. The remainder of the slides are a deeper dive into the technical aspects of the system than TFA gives.

I am not using Ceph but my company makes a system built on an extremely similar architecture and I can vouch for the scalability (modulo implementation stupidities, natch). That being said, our system is difficult to administer due to all the moving parts and its distributed nature - I don't know if Ceph shares this drawback.

[1] https://docs.google.com/viewer?url=http://ceph.newdream.net/...

edit:speling

Interesting, but I have no use for an "experimental" filesystem. I prefer to store my data in filesystems that are stable:

"Although Ceph may not be ready for production environments, it's still useful for evaluation purposes."

It might interest the HN community to know that Sage, the author of Ceph, is one of the founders of DreamHost. After DH became successful he went back to school and got his Ph.D. in distributed storage and built Ceph.