Among the variations the author does recognize, I do wonder how does number of workers influence all of these combinations too: I am convinced just varying them over sync/async category is not sufficient.
I am assuming they did not test for each and every combination and came with exactly the same number, but that they tested one or two configurations to come up with the number.
Another thing to note for real applications, with sync frameworks you are increasing memory demands and memory pressure in a highly loaded system. That probably depends on whether you are hitting memory limits or not (swapping), but it's hard to test for with toy examples.
And yet another is ORM object serialization time: for bulk queries, that can be the most of the response time with Django ORM, SqlAlchemy, SQLObject...
1 comment
[ 3.2 ms ] story [ 12.6 ms ] threadI am assuming they did not test for each and every combination and came with exactly the same number, but that they tested one or two configurations to come up with the number.
Another thing to note for real applications, with sync frameworks you are increasing memory demands and memory pressure in a highly loaded system. That probably depends on whether you are hitting memory limits or not (swapping), but it's hard to test for with toy examples.
And yet another is ORM object serialization time: for bulk queries, that can be the most of the response time with Django ORM, SqlAlchemy, SQLObject...