I use a sharded approach where each room is assigned to a random node.js websocket server when the first person enters it, and it's unloaded when the last person leaves. Just spin up more websocket servers to scale.
We have a hybrid platform with the website running on Ruby on Rails (REE 1.8.7, Unicorn, fronted by Nginx), and the real-time interaction handled by Node.js. I can tell you that the RoR servers fall over LONG before the Node.js servers even begin to break a sweat.
edit: Link to Worlize if you're curious: https://www.worlize.com/ Our Node.JS WebSocket Library: https://github.com/Worlize/WebSocket-Node
How does it compare with Socket.io? I use Socket.io for Review19 - http://review19.com - and am always looking to experiment with new frameworks.. :-)
If you want WebSocket specifically, and don't care about falling back to long-polling techniques, or you want something that allows you to use true binary messages, look at a WebSocket library rather than an abstracted communications library like Socket.IO.
It wouldn't. Designing a system to handle 100k simultaneous users is hard, and a problem few people suffer from.
Why would you ask this question?
Because maybe he is looking to develop a chat system that can scale to 100,000 users? There are many instances where someone is tasked to look at a technology and understand how it scales.
Frankly, it would be helpful for you to tell us how "It wouldn't"
It's a) not a common problem, and b) is obviously non trivial with current technology.
The bandwidth necessary for 100k users would saturate many a pipe, and would immediately require more than one commoditized machine.
It's like saying "That's a cute little Prius you got there. How many semi trailers can it pull?"