Benjamin Marzinski <bmarzins@xxxxxxxxxx> writes: ... [Re: each node exports its disks for a "big share"] > If I'm understanding you correctly, there is no good way that I know of to > do this now. But we are working on changing that. > > right now you can use gnbd or iscsi to export block devices to other nodes. > (I don't know if there are any software iscsi targets that allow you to also > use the exported block device locally. gnbd does. The unh-iscsi target might) > Every node could export a spare partition to all the other nodes. If you had > N nodes, you would have N block devices, all the same size (1 local, and N-1 > remote). If you had cluster aware mirroring, you could set up a mirror on > top of all those devices, and then you would have a consistent image on each > node. However, to make this reliable, you can't have "the big share" as you > called it. If there is information that only exists on one device, It's likely > that the entire cluster would crash in the event of one machine going down. > > As soon as cluster aware mirroring is available (it's being worked on) > This should be easily doable. Huh. I thought that CLVM had mirroring support that works now. In fact, I was planning to test it soon. I plan to set up a GFS filesystem using CLVM on EtherDrive storage blades so that the CLVM-supplied mirroring provides redundancy. -- Ed L Cashin <ecashin@xxxxxxxxxx>
