[
https://issues.apache.org/jira/browse/IMPALA-8759?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16923660#comment-16923660
]
Peter Ebert commented on IMPALA-8759:
-------------------------------------
Yes I am sure that was a thought choice, however even if it's 2x faster with 32
bit vs 64 bit, since it's a single loop at the end of the finalize the cost
would be fixed and doubt it would add even a millisecond, more probably
nanoseconds.
I will say though I don't have much to prove that it is significantly better in
estimation. I did some +very+ basic testing (used tpc-ds generated data) and
although it was more accurate in my very non-scientific tests it was in the 7th
digit and my data set wasn't large enough for that to be significant.
Definitely needs more testing but initially it seems like not a large
difference.
> Use double precision for HLL
> ----------------------------
>
> Key: IMPALA-8759
> URL: https://issues.apache.org/jira/browse/IMPALA-8759
> Project: IMPALA
> Issue Type: Improvement
> Components: Backend
> Affects Versions: Impala 3.2.0
> Reporter: Peter Ebert
> Priority: Major
> Labels: perf, ramp-up
>
> For /be/src/exprs/aggregate-functions-ir.cc the finalize function uses a
> float which is only capable of 6-9 digits of precision. More accurate
> estimates for larger cardinalities (beyond 999,999) should be possible with
> double precision. Another c++ implementation uses double as well
> [https://github.com/dialtr/libcount]
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscr...@impala.apache.org
For additional commands, e-mail: issues-all-h...@impala.apache.org
