perf: Collect Parquet dictionary binary as view #17475
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
coastalwhite
requested review from
ritchie46,
stinodego,
orlp and
c-peters
as code owners
github-actions
bot
added
performance
coastalwhite
force-pushed
the
parquet-binary-views
branch
2 times, most recently
from
d1afbc1
to
23680e4
Compare
This optimizes how Parquet dictionary over binary is collected. Now, instead of
pushing the items one at the time into a buffer. The dictionary is used as a
buffer and views are made into that buffer. This should not only speed up the
Parquet decoder, but should also reduce memory consumption and speed up
subsequent operations.
I did a small benchmark, but this does not really mean much.
```
Benchmark 1: After Optimization
Time (mean ± σ): 2.007 s ± 0.005 s [User: 1.712 s, System: 0.523 s]
Range (min … max): 2.000 s … 2.013 s 10 runs
Benchmark 2: Before Optimization
Time (mean ± σ): 2.285 s ± 0.009 s [User: 1.956 s, System: 0.595 s]
Range (min … max): 2.274 s … 2.306 s 10 runs
Summary
After Optimization ran
1.14 ± 0.01 times faster than Before Optimization
```
coastalwhite
force-pushed
the
parquet-binary-views
branch
from
23680e4
to
faff0ca
Compare
CodSpeed Performance ReportMerging #17475 will improve performances by 26.82%Comparing Summary
Benchmarks breakdown
|
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #17475 +/- ##
==========================================
+ Coverage 80.46% 80.50% +0.03%
==========================================
Files 1483 1483
Lines 194832 195122 +290
Branches 2770 2781 +11
==========================================
+ Hits 156767 157078 +311
+ Misses 37556 37533 -23
- Partials 509 511 +2 ☔ View full report in Codecov by Sentry. |
|
14% on a whole read is quite a lot! Considering you have decompression, decoding etc. |
ritchie46
reviewed
Jul 8, 2024
| let buffer_idx = if max_length <= View::MAX_INLINE_SIZE as usize { | ||
| 0 | ||
| } else { | ||
| values.push_buffer(page_dict.values().clone()) |
There was a problem hiding this comment.
Ah, nice. So we don't have to go through the builder, but immediately use the buffer. 👍
ritchie46
approved these changes
Jul 8, 2024
henryharbeck
pushed a commit
to henryharbeck/polars
that referenced
this pull request
Jul 8, 2024
2 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This optimizes how Parquet dictionary over binary is collected. Now, instead of pushing the items one at the time into a buffer. The dictionary is used as a buffer and views are made into that buffer. This should not only speed up the Parquet decoder, but should also reduce memory consumption and speed up subsequent operations.
I did a small benchmark with the Wikipedia dataset (collect
a.parquetonce), but this does not really mean much.