Originally I implemented the adapter to use single steps and just let it work with erlang's scheduler. I remember I ran into issues where selecting data was actually slower than just batching the results in a list and returning them.
What I would like to do is add a benchee test that works with both methods on a large data set, to really see what sort of improvements can be made and can be quantified.
If we can just utilize step alone and let the scheduler work, that would simplify the NIF code.
Originally I implemented the adapter to use single steps and just let it work with erlang's scheduler. I remember I ran into issues where selecting data was actually slower than just batching the results in a list and returning them.
What I would like to do is add a benchee test that works with both methods on a large data set, to really see what sort of improvements can be made and can be quantified.
If we can just utilize
step
alone and let the scheduler work, that would simplify the NIF code.