While what you said is true in general, I believe it is debatable in this case, as Redis (at least the subset implemented in this experiment) is something very very special, it is essentially a hashmap attached to the network. There aren't much stuff left if you offload the "hashmap" part and the "attached to the network" (I/O + protocol parser) part. In other words, in this case the hashmap implementation itself is the business logic of your application.