Normanu, this is the sort of thing that our company would have to do if someone wanted a petabyte storage array. We'd probably use RDMA for the connections between each of the nodes and the head node (which would have an auto-failover setup with another hot spare head node) in order to provide the necessary throughput. This is the sort of functionality which we'd probably have to contract some linux storage expert for $250/hr to help us engineer this sort of thing. It'd be a lot cheaper and easier and better to have Open-E have this sort of functionality, but it's just not there yet.

And, if you wait two or three years, you could pretty easily get this sort of functionality from a "pNFS" (parallel NFS) setup that I'm sure Open-E will eventually integrate into their storage boxes once it becomes mainstream in the standard linux kernel.

Really, if you want something like you're describing right now, you should look at the GFS2 cluster filesystem. I think you can still use Open-E iSCSI or Fibre Channel targets for the back-end storage. Of course, this probably won't work with Windows.