Who's Online Reviewed
The "who's online" solution we've built scales. However, we focused only briefly on the real reason the solution scales well. Is it because it can serve more than 10,000 requests per second on a single box? Although impressive, that is performance, not scalability.
Scalability means that the general implementation and use of this system will not change if the scope of the problem changes. If the site shrinks, our system becomes less used. If the problem scope increases, to say 100,000 hits per second, we have a solution.
The "who's online" server collects information outside the critical service path. It is passive, and because it uses log streams from mod_log_spread, more can be added without significant overhead. This means that it can scale horizontally. The fact that we could run a "who's online"-enabled spreadlogd instance on every web server without introducing load on any other architectural component is the ultimate testament to its scalability.