I like the survey idea. Would something like an (semi) annual interop event make sense? Maybe co-located with another major conference like SC or ISC, or Interop, or whatever. A survey maybe sets an upper bound on what people are trying (successfully??), an interop test event *maybe* sets a lower bound on what is known to work well. It could point out weaknesses in switching vs memory caching vs random write io latency vs scsi queue depth, etc. Just brainstorming a little here... distributed systems are complex enough, but distributed *storage* just seems to make standard regression testing all the more difficult. The survey to me is a poor mans regression of sorts.