I was making some benchmarks on my Rails application before releasing it in production. When running Apache benchmarks with only 2000 requests and 10 concurrency level I have seen that the last 20% of requests are very long and went from 200 ms to 1600 ms (See the attached image). I believe that it's a configuration issue. What could be wrong? I'm using Apache with Passenger.
What are the best practices of configuring Passenger to serve highest number of requests?