Hi guys,

I am currently looking for a file system that needs to be distributed over several nodes and need redundancy, like RAID5, and the ability to grow if needed, like LVM.
Also not all nodes are going to be located in the same data center, but I guess that's not that important as long as the connection between DCs is sufficiently fast.

I am currently looking at AFS, Coda, GFS and OCFS to see if they have what I need.

Anybody got experience with this? Any suggestions?

Thanks in advance!
Dennis

Edit: I just figured that it may be better to have the RAID5-like setup within one location, and have RAID1-like mirroring between the locations. That would probably better for performance, right?

Initially this just sounds mental. Why different dc's? What is the experience you are after? Why lay one level of raid on another? Using systems like drbd you can keep two block devices identical remotely within reason but that's (afaik) an active standby model only.

As for the distributed volume, you can use AoE to make network block devices and then arrange them in RAID5 or 6 or a nested RAID level, like RAID5+0 or RAID5+1.

If you want to mount this distributed volume R/W in more than one server, then you need to use a shared filesystem one it.
I like OCFS2 for its simplicity.

---------- Post added 19-06-11 at 02:22 ----------

You can use drbd in an active-active model, if you use a shared filesystem like OCFS2 or GFS2