When I heard Rick Boone’s talk at SREcon18 Americas, I was immediately struck by his approach. While capacity planning is really an art, relying partially on past behavior but just as much on intuition, Rick described uncovering the best metric for reliably predicting capacity as needed.
Uber’s services run on their own hardware, and their goal is to always have sufficient capacity without ever having either too much or a shortage that will hurt business. Rick’s approach used machine learning to help pick out the appropriate metric and mathematically predict its impact on a service’s capacity needs. You can watch the video of his talk to learn the approach used. In this interview, Rick discusses why Uber doesn’t use capacity planning and instead relies on capacity engineering.
