Csaba Ringhofer has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/12172 )

Change subject: IMPALA-7979: Enhance decoders to support value-skipping
......................................................................


Patch Set 4: Code-Review+1

(2 comments)

http://gerrit.cloudera.org:8080/#/c/12172/4/be/src/util/rle-test.cc
File be/src/util/rle-test.cc:

http://gerrit.cloudera.org:8080/#/c/12172/4/be/src/util/rle-test.cc@191
PS4, Line 191: values
Is it really useful to print the values? At this point the same length + 
bit_width will lead to the same values, so printing values.size() would be 
enough to reproduce the issue. If the values will also become random, then a 
seed value could be printed.


http://gerrit.cloudera.org:8080/#/c/12172/4/be/src/util/rle-test.cc@275
PS4, Line 275: MakeSequenceBitWidth
It would make sense to do this randomly too, but I am leaning towards creating 
a follow up Jira, which would also deal with adding benchmarks for value 
skipping to benchmarks/rle-benchmark.cc

I do not want to hold up the page index efforts with this, but I am curios 
about the performance of skipping vs materializing for different selectivity 
ratios.

About a possible implementation:
A seed value could be added as parameter, and runs could be added with random 
[1, max_run_length] length until total length is reached. A similar 
implementation already exists in benchmarks/rle-benchmark.cc FillWithRle().



--
To view, visit http://gerrit.cloudera.org:8080/12172
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: Impala-ASF
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: Ib848f1bd71735fe84e8064daf700417b32589f57
Gerrit-Change-Number: 12172
Gerrit-PatchSet: 4
Gerrit-Owner: Zoltan Borok-Nagy <[email protected]>
Gerrit-Reviewer: Csaba Ringhofer <[email protected]>
Gerrit-Reviewer: Impala Public Jenkins <[email protected]>
Gerrit-Reviewer: Tim Armstrong <[email protected]>
Gerrit-Reviewer: Zoltan Borok-Nagy <[email protected]>
Gerrit-Comment-Date: Mon, 21 Jan 2019 10:56:14 +0000
Gerrit-HasComments: Yes

Reply via email to