On Sun, Apr 29, 2012 at 11:22:20PM +0200, Gandalf Corvotempesta wrote: > So, what will you do? RAID1? No raid? RAID10 for write-active filesystems, and RAID6 for archive filesystems. > How does gluster detect a failed disk with no raid? What I don't > understand is how gluster will detect a failure on a disk and the reply > with data on the other server. I'm not sure - that's what the risk is. One would hope that gluster would detect the failed disk and take it out of service, but I see a lot of posts on this list from people who have problems in various failure scenarios (failures to heal and the like). I'm not sure that glusterfs has really got these situations nailed. Indeed, in my experience the gluster client won't even reconnect to a glusterfsd (brick) if the brick has gone away and come back up. You have to manually unmount and remount. That's about the simplest failure scenario you can imagine. > With a raid controller, if controller detect a failure, will reply with > KO to the operating system KO or OK? With a RAID controller (or software RAID), the RAID subsystem should quietly mark the failed drive as unusable and redirect all operations to the working drive. And you will have a way to detect this situation, e.g. /proc/mdstat for Linux software RAID. > Is safer to use a 24 disks server with no raid and with 24 replicated > and distributed bricks (24 on one server and 24 on other server)? In theory they should be the same, and with replicated/distributed you also get the benefit that if an entire server dies, the data remains available. In practice I am not convinced that glusterfs will work well this way.
