If throughput and transaction/analysis speed is a requirement then AWS is not the answer -- anyone who suggests otherwise has never used it for anything larger than a toy dataset. I am currently migrating a large data analysis system (20+ TB running over about 500 EC2 hadoop workers) to a dedicated cluster because the internal EC2 latency reached a tipping point in our analysis runs. If you have the dough to spend on a big vertical system you can spend half as much on a dedicated cluster running a NoSQL solution and probably meet the required spec. The original article was comparing a Porsche Cayenne to a bicycle; the bicycle works for some problems and in certain cases it (or a fleet of them) can solve the problem better, but it was a dishonest strawman comparison for the subject at hand.