Can you discuss how you address the "schema-on-read" problem? One of the downsides of the cheap bit-bucket approach of tech like NoSQL and Hadoop is that it's

Can you discuss how you address the "schema-on-read" problem? One of the downsides of the cheap bit-bucket approach of tech like NoSQL and Hadoop is that it's easy to get the data in, but harder to get it out. The producer of the data has less work. The consumer of the data now has more, since there are no longer guarantees on the structure. An emerging 3rd-wave approach is that of an "eventual schema" (see http://arunxjacob.blogspot.com/2011/11/schema-on-read-not-so-fast.html http://arunxjacob.blogspot.com/2011/11/schema-on-read-not-so... by Arun Jacob, chief of data at Disney). But best practices for eventual schema are immature and evolving. How do you approach the problem of getting data out, and allowing people to use common concepts to query their data?