Having just finished benchmarking a new project I'm writing on nearly all EC2 instance types I have to say nothing beats benchmarking.
Create a spreadsheet, write some benchmarking tools, make sure you aren't actually benchmarking another service the instance depends on (db, API, etc) and go to town. Vary # of request/sec, # of parallel requests, # of app workers/threads/etc.
Write down all your results and then compute amount of 'work' (reqs/sec, computations, db inserts/updates, etc) vs hourly or RI pricing. That should get you some good numbers on what instance type is most cost effective for your workload.
Create a spreadsheet, write some benchmarking tools, make sure you aren't actually benchmarking another service the instance depends on (db, API, etc) and go to town. Vary # of request/sec, # of parallel requests, # of app workers/threads/etc.
Write down all your results and then compute amount of 'work' (reqs/sec, computations, db inserts/updates, etc) vs hourly or RI pricing. That should get you some good numbers on what instance type is most cost effective for your workload.