1 point by edchi 0 minutes ago | edit | delete [-]
We actually tested cuckoo hashing against learned hash-maps and in our setup it was slower. One of the main reasons is that cuckoo hashing is quite sensitive to the payload and we used a scenario were we actually assumed records with meta-data fields as used by many systems at Google. We will update our technical report soon to include those comparisons as well.
However, we do agree with the authors of the blog post, that there are cases in which traditional data structures will remain the best choice and as so often, it always depends. In addition, it might be possible to combine learned models with cuckoo hashing, so those are not even exclusive ideas.
--(from original authors of the Learned Index paper)
We actually tested cuckoo hashing against learned hash-maps and in our setup it was slower. One of the main reasons is that cuckoo hashing is quite sensitive to the payload and we used a scenario were we actually assumed records with meta-data fields as used by many systems at Google. We will update our technical report soon to include those comparisons as well.
However, we do agree with the authors of the blog post, that there are cases in which traditional data structures will remain the best choice and as so often, it always depends. In addition, it might be possible to combine learned models with cuckoo hashing, so those are not even exclusive ideas.
--(from original authors of the Learned Index paper)
We'd love to hear about the size of the payloads used. With the code we posted, larger records take 10-15 more ns per query, which is still faster than the numbers in the paper, but of course YMMV based on the table implementation and other factors.
We actually tested cuckoo hashing against learned hash-maps and in our setup it was slower. One of the main reasons is that cuckoo hashing is quite sensitive to the payload and we used a scenario were we actually assumed records with meta-data fields as used by many systems at Google. We will update our technical report soon to include those comparisons as well. However, we do agree with the authors of the blog post, that there are cases in which traditional data structures will remain the best choice and as so often, it always depends. In addition, it might be possible to combine learned models with cuckoo hashing, so those are not even exclusive ideas.
--(from original authors of the Learned Index paper)