Server side state in http is a wonderful thing to abuse. One adserving system I worked on had a very expensive correlation process where it joined auction stat

Server side state in http is a wonderful thing to abuse. One adserving system I worked on had a very expensive correlation process where it joined auction stats with impression stats. Essentially a big Hadoop job, except run in a stream. Another system I later worked on solved the problem by tying impression requests back to the same server that ran the auction. That server kept the auction data in memory until it either received the impression or until a set amount of time had passed. Hearing how it worked made me feel a bit dirty, then I realized how much money it saved them to do it that way and I was impressed with the simplicity of it.