Definitely a conversation to have with an SE and look at the workload types... how much random read, random write, working set size, sequential read, sequential write. If 6 gigabyte that is a lot of throughput... need to know how many streams, concurrency... how many hosts and how many streams per host... block size. Access to the same data or all different buckets, etc. A lot of factors to help look at sustaining this. A quick sizing at 6GB for all sequential I/O (90% read, 10% write) would be 238x 15K spindles or 462x 7.2K SATA spindles... but I don't trust that without more information. This needs some more analysis and information. More questions than answers for this one...