Speed up the function `min_max_string` #1374

HaoYang670 · 2022-03-01T06:08:34Z

Which issue does this PR close?

Closes #1373 .

Rationale for this change

Improve performance
clean up the code

What changes are included in this PR?

Performance is improved when null_count > 0. See benchmark result.
more functional styles and fewer mutable variables

benchmark

min string 512          time:   [1.3687 us 1.3691 us 1.3696 us]                            
                        change: [-7.7279% -7.5833% -7.4242%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 9 outliers among 100 measurements (9.00%)
  1 (1.00%) high mild
  8 (8.00%) high severe

min nulls string 512    time:   [1.6630 us 1.6647 us 1.6665 us]                                  
                        change: [-33.478% -33.362% -33.244%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  2 (2.00%) high mild
  1 (1.00%) high severe

Are there any user-facing changes?

No

Signed-off-by: remzi <13716567376yh@gmail.com>

HaoYang670 · 2022-03-01T06:19:55Z

Benchmark

Before

min string 512          time:   [1.4824 us 1.4914 us 1.5094 us]                            
                        change: [-2.4834% -0.7235% +0.5457%] (p = 0.44 > 0.05)
                        No change in performance detected.
Found 13 outliers among 100 measurements (13.00%)
  2 (2.00%) high mild
  11 (11.00%) high severe

min nulls string 512    time:   [2.4456 us 2.4528 us 2.4610 us]                                  
                        change: [-0.7299% -0.2597% +0.2043%] (p = 0.26 > 0.05)
                        No change in performance detected.
Found 9 outliers among 100 measurements (9.00%)
  7 (7.00%) high mild
  2 (2.00%) high severe

After

min string 512          time:   [1.5539 us 1.5552 us 1.5565 us]                            
                        change: [-0.4238% -0.2174% -0.0106%] (p = 0.04 < 0.05)
                        Change within noise threshold.
Found 5 outliers among 100 measurements (5.00%)
  1 (1.00%) high mild
  4 (4.00%) high severe

min nulls string 512    time:   [1.8519 us 1.8536 us 1.8552 us]                                  
                        change: [-1.0138% -0.7613% -0.4859%] (p = 0.00 < 0.05)
                        Change within noise threshold.
Found 3 outliers among 100 measurements (3.00%)
  1 (1.00%) low mild
  1 (1.00%) high mild
  1 (1.00%) high severe

codecov-commenter · 2022-03-01T06:25:24Z

Codecov Report

Merging #1374 (4479383) into master (6ee51bc) will increase coverage by 0.04%.
The diff coverage is 62.50%.

@@            Coverage Diff             @@
##           master    #1374      +/-   ##
==========================================
+ Coverage   82.99%   83.03%   +0.04%     
==========================================
  Files         181      181              
  Lines       52933    53003      +70     
==========================================
+ Hits        43932    44013      +81     
+ Misses       9001     8990      -11

Impacted Files	Coverage Δ
arrow/src/compute/kernels/aggregate.rs	`73.22% <62.50%> (-0.11%)`	⬇️
arrow/src/compute/kernels/temporal.rs	`97.22% <0.00%> (-0.13%)`	⬇️
arrow/src/array/transform/mod.rs	`84.52% <0.00%> (ø)`
arrow/src/compute/kernels/comparison.rs	`92.45% <0.00%> (+0.29%)`	⬆️
parquet/src/file/metadata.rs	`94.42% <0.00%> (+0.54%)`	⬆️
arrow/src/record_batch.rs	`94.00% <0.00%> (+2.02%)`	⬆️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6ee51bc...4479383. Read the comment docs.

HaoYang670 · 2022-03-01T06:25:25Z

I have tested that if we replace array.value by array.value_unchecked, there will be 10% extra performance improvement on the benchmark min string 512. But it also introduces an unsafe block.

Signed-off-by: remzi <13716567376yh@gmail.com>

Dandandan · 2022-03-01T08:01:44Z

arrow/src/compute/kernels/aggregate.rs

-                n = item;
-            }
-        }
+        array


According to your benchmark result, this new way a bit slower than before for nulls?

No. You can see the benchmark "min nulls string 512". Time decreases from 2.4us to 1.8us. I this this is performance improvement.

Also for the benchmark "min string 512 ", there is no big performance penalty introduced. (time from 1.49us to 1.55us)

I think we could speed it up a bit more if we could iterate fast on the non-null values.

Ok, let me try the array.value_unchecked()

10% extra speed up Signed-off-by: remzi <13716567376yh@gmail.com>

HaoYang670 · 2022-03-03T02:00:25Z

min string 512          time:   [1.3687 us 1.3691 us 1.3696 us]                            
                        change: [-7.7279% -7.5833% -7.4242%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 9 outliers among 100 measurements (9.00%)
  1 (1.00%) high mild
  8 (8.00%) high severe

min nulls string 512    time:   [1.6630 us 1.6647 us 1.6665 us]                                  
                        change: [-33.478% -33.362% -33.244%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  2 (2.00%) high mild
  1 (1.00%) high severe

Using value_unchecked can get 8% extra speedup for not-null values

Signed-off-by: remzi <13716567376yh@gmail.com>

alamb · 2022-03-03T17:47:39Z

Thanks @HaoYang670 and @Dandandan

HaoYang670 added 2 commits March 1, 2022 12:08

clean up the code

e938ffb

Signed-off-by: remzi <13716567376yh@gmail.com>

bring back the optimization when null count is zero

ad64d98

Signed-off-by: remzi <13716567376yh@gmail.com>

github-actions bot added the arrow Changes to the arrow crate label Mar 1, 2022

pretty the trait bound and update comment

89d99d3

Signed-off-by: remzi <13716567376yh@gmail.com>

Dandandan reviewed Mar 1, 2022

View reviewed changes

Dandandan approved these changes Mar 2, 2022

View reviewed changes

use value_unchecked to replace array.value

a5ae1ef

10% extra speed up Signed-off-by: remzi <13716567376yh@gmail.com>

update the performance data

4479383

Signed-off-by: remzi <13716567376yh@gmail.com>

alamb changed the title ~~Speed up the function min_max_string~~ Improve performance of min and max aggregation kernls (optimize min_max_string) Mar 3, 2022

alamb changed the title ~~Improve performance of min and max aggregation kernls (optimize min_max_string)~~ Improve performance of min and max aggregation kernels without nulls Mar 3, 2022

alamb changed the title ~~Improve performance of min and max aggregation kernels without nulls~~ Speed up the function min_max_string Mar 3, 2022

alamb merged commit c947027 into apache:master Mar 3, 2022

HaoYang670 deleted the refactor_min_max_string branch March 4, 2022 00:55

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Speed up the function `min_max_string` #1374

Speed up the function `min_max_string` #1374

HaoYang670 commented Mar 1, 2022 •

edited

Loading

HaoYang670 commented Mar 1, 2022

codecov-commenter commented Mar 1, 2022 •

edited

Loading

HaoYang670 commented Mar 1, 2022

Dandandan Mar 1, 2022

HaoYang670 Mar 1, 2022 •

edited

Loading

HaoYang670 Mar 1, 2022 •

edited

Loading

HaoYang670 Mar 2, 2022

Dandandan Mar 2, 2022

HaoYang670 Mar 2, 2022

HaoYang670 commented Mar 3, 2022 •

edited

Loading

alamb commented Mar 3, 2022

Speed up the function min_max_string #1374

Speed up the function min_max_string #1374

Conversation

HaoYang670 commented Mar 1, 2022 • edited Loading

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

benchmark

Are there any user-facing changes?

HaoYang670 commented Mar 1, 2022

Benchmark

Before

After

codecov-commenter commented Mar 1, 2022 • edited Loading

Codecov Report

HaoYang670 commented Mar 1, 2022

Dandandan Mar 1, 2022

Choose a reason for hiding this comment

HaoYang670 Mar 1, 2022 • edited Loading

Choose a reason for hiding this comment

HaoYang670 Mar 1, 2022 • edited Loading

Choose a reason for hiding this comment

HaoYang670 Mar 2, 2022

Choose a reason for hiding this comment

Dandandan Mar 2, 2022

Choose a reason for hiding this comment

HaoYang670 Mar 2, 2022

Choose a reason for hiding this comment

HaoYang670 commented Mar 3, 2022 • edited Loading

alamb commented Mar 3, 2022

Speed up the function `min_max_string` #1374

Speed up the function `min_max_string` #1374

HaoYang670 commented Mar 1, 2022 •

edited

Loading

codecov-commenter commented Mar 1, 2022 •

edited

Loading

HaoYang670 Mar 1, 2022 •

edited

Loading

HaoYang670 Mar 1, 2022 •

edited

Loading

HaoYang670 commented Mar 3, 2022 •

edited

Loading