[krg]

Round 1 used Passenger, but the feedback we got was that Unicorn performed better so we switched to that. Currently, we're aiming to show each framework in the best possible production configuration. In the future we plan to show multiple web/app servers per framework so that you could compare Ruby on Passenger vs. Ruby on Unicorn.

[FooBarWidget]

There is a simple explanation for that. Phusion Passenger always proxies data from the web server to another process, for stability and security reasons. If you benchmark Unicorn directly, without putting it behind a reverse proxy, Unicorn will look faster simply because you're avoiding another kernel socket operation.

However as I explained in https://news.ycombinator.com/item?id=5727232, Unicorn is always supposed to be put behind a reverse proxy. If you do that you should find different results.

Also, there's a lot of tuning in Phusion Passenger that can help performance. The default is optimized for usability and stability. For example if you don't prespawn processes, and let Phusion Passenger spawn them on the first request, you'll be adding tens of seconds to the benchmark time, which would greatly disadvantage Phusion Passenger in an unfair manner. You should set at least:

passenger_min_instances

passenger_max_pool_size

passenger_pre_start

[_mikz]

Passenger has some config options to spawn more processes and help the load. But the results will not be much different from unicorn/thin/whatever. Maybe in units of %.