How to model or justify Non Functional Requirements in Architecture document ? For example, the NFR is accepted response time is x secs. How to justify and ensure the final product will meet this requirement? Any pointer to good reference would be great.
I would put a hook at the entry point and exit point (think InterceptingFilter) to log the message arrival time/response time to ensure that the response time is always within acceptable NFR limits. It is kind of difficult IMHO to put a number to the TPS for a specific hardware configuration unless you have tested a very similar configuration.
It's quite tricky question. I think they do not ask how to measure this rate, but they ask a prove that your servers will provide needed timings. The only way is to operate with server resource numbers. e.g. "User session takes in average X msecs of computing resources, so processor with Y MHz will server Z simultaneous client with needed response time"
I believe that you cannot prove any numbers until you are prototyping, test and measure in an environment close to the production one.
Most of assignments doesn't require more than 1k concurrent users and if you design your solution to scale there are good chances to reach the NFR numbers by adding more hardware if necessary.