This article drives at what may be the single biggest factor to consider when building a system to scale: sharding. Nowadays, every time I think about an opera

This article drives at what may be the single biggest factor to consider when building a system to scale: sharding. Nowadays, every time I think about an operation that the user will have to wait for, I consider if it "jumps the shard". If so, it'll ultimately have to fire off a request to a separate server, and perhaps do an expensive join. It'll be slow. It had damn well better be worth it.