As engineers we are remarkably badly trained in thinking about cost. I don’t mean some big O order of magnitude cost but real dollar cost. You could given me five fault-tolerant algorithms and almost with my eyes closed I could pick the best one. But if you would then ask me what the difference would be in cost between the best and second-best when the system would start to scale I would falter.

I learned how to incorporate cost into architecture the hard way. At Amazon.com the margins are, as in most retail, razor thin and as such any fluctuation…