Today I was trying to explain the differences between threaded and evented web servers to a friend on a high level. Usually the threaded model is explained with a grocery store analogy and I started to wonder how that analogy could be extended to explain the evented model too.
Threaded model
Lets revise the grocery store analogy for the threaded web servers first. Grocery store is the web server, cashiers are the threads and customers are the incoming requests.
During a late afternoon when everybody is leaving work and going for grocery shopping the store will be filled with customers. For the store to work properly the store owner has to hire multiple cashiers to cash the customers concurrently in order to keep them from waiting in long lines. Web servers spawn multiple threads to serve requests concurrently to keep users from waiting pages to load. Hiring cashiers cost money as spawning threads cost memory and CPU cycles.
Evented model
In the evented model the web server has only a single thread. How a store with single cashier could function in a rush hour?
The cashiers use most of the their time scanning bar codes as threads use most of the their time waiting for I/O in threaded web servers. In some modern stores the customers scan the products themselves while picking them to their baskets. This means that there is no limit how many products can be scanned concurrently. Same goes for evented web servers. The thread is not waiting for any I/O. It just does the real computing work as the cashier does only cashing of the customers. This is very efficient.
This is of course has some drawbacks. A single troublesome customer can block the entire checkout line for all the customers. This is true for evented web servers too. It will go to bankruptcy if Mr. Fibonacci is let in to the store to occupy the only cashier with long computations. The single thread cannot respond to any other requests if it is busy computing something.
Luckily this is not usually the case with web servers and if the web application really needs to do some intensive computations the work must be handed to some other worker process. This way the thread can keep responding to the requests in a timely fashion. Like in the store, the cashier cannot go of doing something else for long periods of time without creating long lines. He will need someone to help on those tasks.
Evented systems can be more efficient, but you have to take greater care of what you let run in it.
Conlusion
While this is not a perfect analogy, but I do like how this explains some of the drawbacks too. Some evented systems (read Node.js) get way too much hype for just being fast without considering the gotchas and the real benefits they bring in, but that’s a whole another blog post.
I hope this helps some others to grasp the evented model too.