On 03/03/2010 01:12 PM, Tejas N. Bhise wrote: > Congrats !! .. you have been converted to glusterfs too :-) .. hehe. Well, we're still testing and we plan to look at Lustre next. Though I must say we really liked GlusterFS and are secretely rooting for it :) > The backend subvolumes can be just directories. They need not be full > filesystems/partitions etc. So on one machine, you can create one big > filesystem and make directories in it. Each directory can be a > backend subvolume. This is thin provisioning in GlusterFS. Then you > need not bother about smallest disk. You can delete files from one > directory ( hence one backend volume ) to create space in another > directory ( hence another backend volume ). As you play with the > system, you will see more and more possibilities :-). Hum! Interesting. Thanks for the tip. I'm still curious about why the previous configuration didn't work. Can you tell me if using the replication translator over distributed volumes is supported? > So what purpose do you intend to put the system to ? Well, we have a driving project, an application that needs scalable storage, and we'd like to use the chosen system for other existing projects that are now running over NFS exported filesystems (hence the need for posix) and for future projects with largish/scalable storage needs. We plan on trying it for storing our VM images too, which I've seen that some people do. So, basically, we need a scalable, redundant and flexible storage system, hopefully also easy to operate and configure when adding new nodes and disks. GlusterFS seemed perfect, with its layered approach and shared-nothing design :) Thank you all very much for your work! regards, z?
