[shadeless]

IIRC there were some issues in how Phoenix was configured for the techempower benchmarks, I expect that it will have much better results in the next one.

You may want to look at these benchmarks also:

https://github.com/mroth/phoenix-showdown

https://gist.github.com/omnibs/e5e72b31e6bd25caf39a

Here is a post about using 2 million active websockets connections in Phoenix: http://www.phoenixframework.org/blog/the-road-to-2-million-w...

From my experience, Rails and Phoenix speed isn't comparable. Even in development mode (with code-reloading etc) Phoenix is much faster than a production Rails server. I've also replaced three Node.js servers with one Elixir server which did the same work with less memory and CPU usage, but I'm aware that most rewrites have a similar story, so I won't claim that this is enough proof.