Use 1024-value pco pages in btrblocks float/integer schemes#7922
Closed
joseph-isaacs wants to merge 2 commits into
Closed
Use 1024-value pco pages in btrblocks float/integer schemes#7922joseph-isaacs wants to merge 2 commits into
joseph-isaacs wants to merge 2 commits into
Conversation
Adds a divan benchmark that compares pcodec with the default page size against pcodec configured with a 1024-value page size on f64 and i64 data. The bench reports the compression ratio of each variant on startup and times compression, full decompression, and per-element scalar_at access for both, so we can quantify the size/random-access tradeoff of smaller pco pages. Signed-off-by: Claude <noreply@anthropic.com>
The btrblocks PcoScheme for floats and integers compressed with an 8192- value page size. Shrinking the page to 1024 keeps the per-row size essentially unchanged (~0.03 bits/value for f64, ~0.07 bits/value for i64 in the new pcodec_page_size bench) while shrinking the decode window 8x for random access via scalar_at. Signed-off-by: Claude <noreply@anthropic.com>
Merging this PR will improve performance by 20.33%
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| 🆕 | Simulation | pcodec_compress_i64_1k_page |
N/A | 9.7 ms | N/A |
| 🆕 | Simulation | pcodec_compress_i64_default_page |
N/A | 10.6 ms | N/A |
| 🆕 | Simulation | pcodec_decompress_f64_1k_page |
N/A | 3.2 ms | N/A |
| 🆕 | Simulation | pcodec_decompress_f64_default_page |
N/A | 2.3 ms | N/A |
| 🆕 | Simulation | pcodec_compress_f64_1k_page |
N/A | 10.6 ms | N/A |
| 🆕 | Simulation | pcodec_compress_f64_default_page |
N/A | 11.2 ms | N/A |
| 🆕 | Simulation | pcodec_scalar_at_i64_default_page |
N/A | 1.4 s | N/A |
| 🆕 | Simulation | pcodec_scalar_at_f64_1k_page |
N/A | 34.4 ms | N/A |
| 🆕 | Simulation | pcodec_scalar_at_f64_default_page |
N/A | 1.6 s | N/A |
| 🆕 | Simulation | pcodec_scalar_at_i64_1k_page |
N/A | 32.3 ms | N/A |
| ⚡ | Simulation | new_bp_prim_test_between[i16, 32768] |
134.1 µs | 120.2 µs | +11.64% |
| ⚡ | Simulation | new_bp_prim_test_between[i32, 32768] |
169.9 µs | 141 µs | +20.51% |
| ⚡ | Simulation | new_bp_prim_test_between[i32, 16384] |
109.1 µs | 94.7 µs | +15.21% |
| ⚡ | Simulation | new_bp_prim_test_between[i64, 16384] |
144.4 µs | 115.1 µs | +25.52% |
| ⚡ | Simulation | new_bp_prim_test_between[i64, 32768] |
236.7 µs | 177.9 µs | +33.03% |
| ⚡ | Simulation | new_alp_prim_test_between[f64, 16384] |
148.8 µs | 126.9 µs | +17.31% |
Tip
Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.
Comparing claude/pcodec-1k-page-benchmark-mc8sI (617f690) with develop (7349cd6)
Footnotes
-
24 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩
joseph-isaacs
added
the
action/benchmark
github-actions
Bot
removed
the
action/benchmark
Contributor
Polar Signals Profiling ResultsLatest Run
Powered by Polar Signals Cloud |
Contributor
Benchmarks: PolarSignals ProfilingVortex (geomean): 1.008x ➖ datafusion / vortex-file-compressed (1.008x ➖, 0↑ 0↓)
|
Contributor
File Sizes: PolarSignals ProfilingNo file size changes detected. |
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.989x ➖, 1↑ 0↓)
datafusion / vortex-compact (1.018x ➖, 0↑ 1↓)
datafusion / parquet (1.014x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (0.968x ➖, 1↑ 0↓)
duckdb / vortex-compact (1.004x ➖, 0↑ 0↓)
duckdb / parquet (0.992x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: FineWeb NVMeFile Size Changes (1 files changed, +0.0% overall, 1↑ 0↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.970x ➖, 4↑ 1↓)
datafusion / vortex-compact (1.142x ❌, 0↑ 15↓)
datafusion / parquet (1.039x ➖, 0↑ 2↓)
datafusion / arrow (0.890x ✅, 12↑ 3↓)
duckdb / vortex-file-compressed (0.991x ➖, 1↑ 1↓)
duckdb / vortex-compact (1.165x ❌, 0↑ 13↓)
duckdb / parquet (0.978x ➖, 2↑ 0↓)
duckdb / duckdb (0.958x ➖, 4↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-H SF=1 on NVMEFile Size Changes (7 files changed, +0.5% overall, 7↑ 0↓)
Totals:
|
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.011x ➖, 0↑ 1↓)
datafusion / vortex-compact (1.005x ➖, 0↑ 2↓)
datafusion / parquet (1.016x ➖, 1↑ 1↓)
duckdb / vortex-file-compressed (1.039x ➖, 0↑ 8↓)
duckdb / vortex-compact (1.070x ➖, 1↑ 30↓)
duckdb / parquet (1.006x ➖, 0↑ 0↓)
duckdb / duckdb (1.034x ➖, 0↑ 5↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-DS SF=1 on NVMEFile Size Changes (15 files changed, +0.8% overall, 15↑ 0↓)
Totals:
|
Contributor
Benchmarks: FineWeb S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.156x ➖, 0↑ 2↓)
datafusion / vortex-compact (0.948x ➖, 1↑ 0↓)
datafusion / parquet (1.247x ➖, 0↑ 3↓)
duckdb / vortex-file-compressed (1.042x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.044x ➖, 0↑ 0↓)
duckdb / parquet (1.052x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
🚨🚨🚨❌❌❌ SQL BENCHMARK FAILED ❌❌❌🚨🚨🚨Benchmark |
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) duckdb / vortex-file-compressed (0.980x ➖, 1↑ 0↓)
duckdb / vortex-compact (1.015x ➖, 0↑ 1↓)
duckdb / parquet (0.994x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: Statistical and Population GeneticsFile Size Changes (1 files changed, +1.5% overall, 1↑ 0↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=10 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.026x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.104x ❌, 0↑ 9↓)
datafusion / parquet (1.031x ➖, 0↑ 0↓)
datafusion / arrow (0.960x ➖, 4↑ 0↓)
duckdb / vortex-file-compressed (1.026x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.125x ❌, 0↑ 15↓)
duckdb / parquet (1.008x ➖, 0↑ 0↓)
duckdb / duckdb (1.011x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
File Sizes: TPC-H SF=10 on NVMEFile Size Changes (22 files changed, +0.5% overall, 22↑ 0↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.140x ➖, 0↑ 5↓)
datafusion / vortex-compact (1.053x ➖, 0↑ 1↓)
datafusion / parquet (1.057x ➖, 0↑ 3↓)
duckdb / vortex-file-compressed (0.994x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.993x ➖, 0↑ 0↓)
duckdb / parquet (1.043x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: Random AccessVortex (geomean): 1.006x ➖ unknown / unknown (1.003x ➖, 0↑ 1↓)
|
Contributor
Benchmarks: CompressionVortex (geomean): 0.998x ➖ unknown / unknown (0.999x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: TPC-H SF=10 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (1.080x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.222x ➖, 0↑ 6↓)
datafusion / parquet (1.104x ➖, 0↑ 2↓)
duckdb / vortex-file-compressed (1.178x ➖, 0↑ 2↓)
duckdb / vortex-compact (1.145x ➖, 0↑ 1↓)
duckdb / parquet (1.078x ➖, 0↑ 0↓)
Full attributed analysis
|
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
Summary
vortex-btrblocks'PcoSchemefor floats and integers configured pcodecwith an 8192-value page size. Drop that to 1024 so random access through
scalar_atdecodes an 8x smaller window.vortex/benches/pcodec_page_size.rs— a divan bench that comparespcodec with the default page size against pcodec configured with 1024-value
pages, on
f64andi64data. It reports compression ratios at startup andtimes compression, full decompression, and
scalar_atfor both variants.Why 1024
Sample output of the new bench (100K values, level 3):
The per-row overhead of 1k pages is small (~0.03 bits/value f64, ~0.07
bits/value i64) and
scalar_atonly has to decode 1024 values instead of afull chunk.
Test plan
cargo clippy -p vortex-btrblocks --all-features --all-targetscargo test -p vortex-btrblocks— 35 unit tests + 3 doctests passcargo bench -p vortex --bench pcodec_page_size --no-run— bench buildspcodec_page_size --test— all 10 bench cases executehttps://claude.ai/code/session_01Hr3v3EaG9XjKypGyyq2BWF
Generated by Claude Code