On Wed, Feb 29, 2012 at 11:59 AM, Michael ODonnell < [email protected]> wrote:
> However, being a hot copy means any > filesystem metadata would be in whatever state the failed node left > them, which may or may not be an acceptable risk in some situations. > Some sort of distributed filesystem (on redundant storage) shared among > all the machines in question seems preferable... > Except in my case, I'm looking for distributed block devices on which my VMs will put their own file system. I could use image files on a distributed file system (which is how I tried to use gluster last year), but then I end up with the terrible overhead of a file system inside another file system. This tends to create a ridiculous amount of head seeks and quite a bit more IO than a single file system, even with noatime,nodiratime set all around. With atime on both file systems, IO goes strait to pot.
_______________________________________________ gnhlug-discuss mailing list [email protected] http://mail.gnhlug.org/mailman/listinfo/gnhlug-discuss/
