Aleksandr Efimov has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/24896 )

Change subject: IMPALA-15381: Reduce reservoir RNG state
......................................................................


Patch Set 3:

Thanks for trying it. The gain is per group, not per row, so HistogramTest (7 
groups at the 20000-sample cap) is exactly the case where nothing changes. Lab 
cluster, 3 nodes, tpch_parquet.lineitem SF1, debug build, so the times are only 
good relative to each other:

appx_median(l_shipmode) ... group by l_orderkey, 1.5M groups of ~4 rows: 4653 
-> 598 bytes per group outside the reservation, 5.32GB -> 346MB of intermediate 
rows into the merge, and 5.3-5.9s -> 1.71-1.76s per query. Grouped by 
l_shipmode instead, 7 groups at the cap: unchanged, like HistogramTest.

mt19937_64 seeds 312 words per group in the constructor, and the whole state 
goes into every serialized intermediate row.


--
To view, visit http://gerrit.cloudera.org:8080/24896
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I9e046e3f0c1b91f1761ce578601a330bd864eb84
Gerrit-Change-Number: 24896
Gerrit-PatchSet: 3
Gerrit-Owner: Aleksandr Efimov <[email protected]>
Gerrit-Reviewer: Aleksandr Efimov <[email protected]>
Gerrit-Reviewer: Csaba Ringhofer <[email protected]>
Gerrit-Reviewer: Impala Public Jenkins <[email protected]>
Gerrit-Reviewer: Joe McDonnell <[email protected]>
Gerrit-Reviewer: Michael Smith <[email protected]>
Gerrit-Reviewer: Riza Suminto <[email protected]>
Gerrit-Comment-Date: Tue, 22 Sep 2026 18:11:07 +0000
Gerrit-HasComments: No

Reply via email to