Re:What are Bloom filters? (2015)
by Retric on Friday August 05, 2016 @07:18PM
(#12236328)
In your approach comments have votes the server must check the bloom filter for each comment then do a fetch for each match in the case of 500 votes that's 500 fetches which are at least votes * (log comments votes) because you can't trust the filter.
My approach that's client side most of the time. Further it's O (n) to compare matches between two sorted lists so just sort them in O (n log n). Further log 500 is faster than most hashes.
And of course for the most common case of no votes it's a single lookup and done.
PS: As to changing the back end, yes if the initial implementation is bad then you might have a point. But, again fixing the problem also works and produces far less code to maintain.
My approach that's client side most of the time. Further it's O (n) to compare matches between two sorted lists so just sort them in O (n log n). Further log 500 is faster than most hashes.
And of course for the most common case of no votes it's a single lookup and done.
PS: As to changing the back end, yes if the initial implementation is bad then you might have a point. But, again fixing the problem also works and produces far less code to maintain.