This is a demo of the new ZRANGEBYLEX Redis command. It autocompletes the query field using all the 8 millions of unique lines of the Linux kernel source code as autocompletion dictionary.
Every time the user types something, the page sends a request to a small PHP snippet called search.php which sends a query to Redis, and returns the result to the web page as a JSON array.
As you type in the input field below, you'll see the list of suggestions appearing.
The Redis instance uses 1 GB of memory to hold the 8 million unique lines into a sorted set. The search.php script (source code in this Gist) is not very optimized and creates a new Redis connection at every call, but this does not add enough latency for the user to notice. In practice we can auto-complete thousands of simultaneous connections (that likely translates to hundreds of thousands real users, as a connection is created only when the user stops typing something, and immediately destroyed) in real time, with a delay that is mostly the round-trip-time between the client and the web server.
ETA: The new ZRANGEBYLEX command will be available in Redis 2.8.9 before the end of April 2014. It is already available in the 2.8 branch at Github.