flask-profiler
flask-profiler copied to clipboard
muatik
besides mean average, mode metric should also be available
sometimes a few outliner elapsed values can dramatically change average value and this may lead to wrong evaluation. to decrease this side effect, mode calculation should be available.
How many decimal places should be considered in your opinion for mode calculations?
I would try to write an API for the same.
Looking at request times within a few projects, it looks like 6 decimal places would be sufficient.
But I am not convinced yet that mode calculation will be useful. On the other hand, I am also thinking of calculating standard deviation to devise a better way for elimination of outliners. For this; 1.step: calculate standard deviation 2.step: ignore numbers which are out of standard deviation 3.step: calculate mean of request times which are in standard deviation.
Check this out: https://perfwork.wordpress.com/2012/04/09/response-time-metric-for-sla/ Percentiles are the solution. We can go for 90% and a 20% too.