Experiment: optimize compressor selection for scan time#8780
Closed
connortsui20 wants to merge 1 commit into
Closed
Experiment: optimize compressor selection for scan time#8780connortsui20 wants to merge 1 commit into
connortsui20 wants to merge 1 commit into
Conversation
connortsui20
added
experimental
connortsui20
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): 0.912x ➖ How to read Verdict and Engines
datafusion / vortex-file-compressed (0.912x ➖, 6↑ 0↓)
File Size Changes (1 files changed, +0.2% overall, 1↑ 0↓)
Totals:
|
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.943x ➖, 3↑ 3↓)
datafusion / vortex-compact (1.016x ➖, 1↑ 2↓)
datafusion / parquet (1.010x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (0.908x ➖, 3↑ 0↓)
duckdb / vortex-compact (1.119x ❌, 0↑ 1↓)
duckdb / parquet (1.017x ➖, 0↑ 0↓)
File Size Changes (2 files changed, +13.8% overall, 2↑ 0↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.053x ➖, 0↑ 2↓)
datafusion / vortex-compact (1.028x ➖, 0↑ 0↓)
datafusion / parquet (1.050x ➖, 1↑ 6↓)
datafusion / arrow (1.071x ➖, 0↑ 3↓)
duckdb / vortex-file-compressed (1.045x ➖, 0↑ 1↓)
duckdb / vortex-compact (1.000x ➖, 2↑ 2↓)
duckdb / parquet (1.057x ➖, 0↑ 3↓)
duckdb / duckdb (1.052x ➖, 0↑ 2↓)
File Size Changes (18 files changed, +8.9% overall, 11↑ 7↓)
Totals:
|
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.965x ➖, 21↑ 3↓)
datafusion / vortex-compact (0.942x ➖, 31↑ 8↓)
datafusion / parquet (0.948x ➖, 22↑ 0↓)
duckdb / vortex-file-compressed (1.014x ➖, 4↑ 10↓)
duckdb / vortex-compact (0.810x ✅, 67↑ 1↓)
duckdb / parquet (1.021x ➖, 0↑ 2↓)
duckdb / duckdb (0.990x ➖, 0↑ 0↓)
File Size Changes (48 files changed, +12.2% overall, 22↑ 26↓)
Totals:
|
Contributor
Benchmarks: FineWeb S3Verdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.075x ➖, 0↑ 1↓)
datafusion / vortex-compact (1.133x ➖, 0↑ 2↓)
datafusion / parquet (1.012x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.145x ➖, 0↑ 1↓)
duckdb / vortex-compact (0.950x ➖, 0↑ 0↓)
duckdb / parquet (0.986x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) How to read Verdict and Engines
duckdb / vortex-file-compressed (0.973x ➖, 2↑ 0↓)
duckdb / vortex-compact (0.904x ➖, 1↑ 2↓)
duckdb / parquet (0.988x ➖, 0↑ 0↓)
File Size Changes (2 files changed, +13.9% overall, 2↑ 0↓)
Totals:
|
Contributor
Benchmarks: Clickbench Sorted on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.032x ➖, 1↑ 3↓)
datafusion / parquet (1.007x ➖, 0↑ 0↓)
duckdb / vortex-file-compressed (1.059x ➖, 1↑ 3↓)
duckdb / parquet (0.999x ➖, 0↑ 0↓)
duckdb / duckdb (1.016x ➖, 0↑ 0↓)
File Size Changes (201 files changed, +16.3% overall, 167↑ 34↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=10 on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.995x ➖, 0↑ 0↓)
datafusion / vortex-compact (0.950x ➖, 5↑ 0↓)
datafusion / parquet (1.020x ➖, 0↑ 1↓)
datafusion / arrow (1.061x ➖, 0↑ 5↓)
duckdb / vortex-file-compressed (0.986x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.021x ➖, 3↑ 6↓)
duckdb / parquet (0.987x ➖, 1↑ 0↓)
duckdb / duckdb (1.012x ➖, 0↑ 0↓)
File Size Changes (48 files changed, +8.9% overall, 39↑ 9↓)
Totals:
|
Contributor
Benchmarks: Clickbench on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (1.024x ➖, 1↑ 6↓)
datafusion / parquet (1.045x ➖, 0↑ 9↓)
duckdb / vortex-file-compressed (0.982x ➖, 2↑ 1↓)
duckdb / parquet (0.999x ➖, 0↑ 0↓)
duckdb / duckdb (0.997x ➖, 0↑ 0↓)
File Size Changes (201 files changed, +24.2% overall, 199↑ 2↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.875x ➖, 3↑ 2↓)
datafusion / vortex-compact (0.911x ➖, 0↑ 0↓)
datafusion / parquet (1.075x ➖, 0↑ 5↓)
duckdb / vortex-file-compressed (0.972x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.972x ➖, 0↑ 0↓)
duckdb / parquet (0.978x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: Random AccessVortex (geomean): 1.044x ➖ How to read Verdict and Engines
unknown / unknown (1.030x ➖, 1↑ 2↓)
|
Contributor
Benchmarks: Appian on NVMEVerdict: No clear signal (low confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.917x ➖, 4↑ 0↓)
datafusion / parquet (0.943x ➖, 2↑ 0↓)
duckdb / vortex-file-compressed (0.950x ➖, 2↑ 0↓)
duckdb / parquet (0.987x ➖, 0↑ 0↓)
duckdb / duckdb (0.961x ➖, 1↑ 0↓)
File Size Changes (19 files changed, +6.8% overall, 3↑ 16↓)
Totals:
|
Contributor
Benchmarks: TPC-H SF=10 on S3Verdict: No clear signal (environment too noisy confidence) How to read Verdict and Engines
datafusion / vortex-file-compressed (0.969x ➖, 0↑ 0↓)
datafusion / vortex-compact (0.946x ➖, 1↑ 1↓)
datafusion / parquet (0.943x ➖, 1↑ 0↓)
duckdb / vortex-file-compressed (0.967x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.954x ➖, 0↑ 0↓)
duckdb / parquet (0.952x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: CompressionVortex (geomean): 1.037x ➖ How to read Verdict and Engines
unknown / unknown (1.025x ➖, 3↑ 11↓)
|
connortsui20
added
action/benchmark
connortsui20
marked this pull request as ready for review
July 16, 2026 15:43
connortsui20
removed
the
action/benchmark
connortsui20
added
the
action/benchmark
connortsui20
marked this pull request as draft
July 16, 2026 15:45
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 521d01e2fb
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Comment on lines
580
to
+584
| if [ "${{ inputs.mode }}" != "pr" ]; then | ||
| packages+=(--bin lance-bench) | ||
| features+=(--features unstable_encodings) | ||
| fi | ||
| cargo build "${packages[@]}" --profile release_debug --features unstable_encodings | ||
| cargo build "${packages[@]}" "${features[@]}" --profile release_debug |
There was a problem hiding this comment.
When inputs.mode == 'pr', features stays empty here, but this workflow's compare step still downloads data.json.gz from the develop benchmark results, and develop mode takes this same branch the other way and adds --features unstable_encodings. That means PR SQL benchmark comments now compare stable-encoding PR binaries against unstable-encoding develop baselines, so unrelated PRs can be reported as faster/slower because of the feature-policy difference rather than their changes. Please either produce a matching stable baseline or keep the PR and develop feature sets aligned for comparisons.
Useful? React with 👍 / 👎.
| /// [`SizeCost`] because the scan calibration has not been validated for those schemes. | ||
| /// [`ReadTimeCost`](vortex_compressor::cost::ReadTimeCost) is available for deterministic | ||
| /// sequential-scan objectives that combine estimated I/O and decode time. | ||
| pub fn with_cost_model(mut self, cost_model: Arc<dyn CostModel>) -> Self { |
There was a problem hiding this comment.
This new public hook lets downstream users pass a custom CostModel, but a crate that depends only on vortex-btrblocks or vortex::compressor cannot implement that trait because CostModel::cost takes Candidate<'_> and neither facade re-exports Candidate. In that scenario the method signature has no usable public path without adding a lower-level vortex-compressor dependency, which undercuts the custom model API; please re-export Candidate alongside CostModel through the same public modules.
Useful? React with 👍 / 👎.
Member
Author
There was a problem hiding this comment.
ok yeah please make that change for me
connortsui20
force-pushed
the
codex/cost-model-query-performance
branch
from
July 16, 2026 15:53
521d01e to
a1f9c13
Compare
connortsui20
added
action/benchmark
connortsui20
force-pushed
the
ct/compressor-cost-model
branch
from
July 16, 2026 15:54
cb878e4 to
43f6c42
Compare
Signed-off-by: Connor Tsui <connor.tsui20@gmail.com>
connortsui20
force-pushed
the
codex/cost-model-query-performance
branch
from
July 16, 2026 15:55
a1f9c13 to
4e73ee6
Compare
connortsui20
added
action/benchmark-sql-full
Merging this PR will improve performance by 59.92%
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ⚡ | WallTime | cuda/bitpacked_u8/unpack/3bw[100M] |
646.3 µs | 299.2 µs | ×2.2 |
| ⚡ | WallTime | cuda/bitpacked_patched_u8/unpack/1%[100M] |
768.9 µs | 380.4 µs | ×2 |
| ⚡ | WallTime | cuda/bitpacked_patched_u8/unpack/10%[100M] |
828.6 µs | 479.9 µs | +72.64% |
| ⚡ | WallTime | cuda/bitpacked_patched_u16/unpack/1%[100M] |
712.3 µs | 594.9 µs | +19.74% |
| ⚡ | WallTime | cuda/bitpacked_patched_u16/unpack/10%[100M] |
831 µs | 717.2 µs | +15.87% |
| 🆕 | Simulation | bitpacked_decode |
N/A | 2.6 ms | N/A |
| 🆕 | Simulation | bitpacked_eq_fill |
N/A | 2.4 ms | N/A |
| 🆕 | Simulation | bitpacked_filter_10pct |
N/A | 3.7 ms | N/A |
| 🆕 | Simulation | dict_decode |
N/A | 39.3 ms | N/A |
| 🆕 | Simulation | dict_eq |
N/A | 4.5 ms | N/A |
| 🆕 | Simulation | dict_like |
N/A | 62.4 ms | N/A |
| 🆕 | Simulation | fsst_decode |
N/A | 28.2 ms | N/A |
| 🆕 | Simulation | fsst_eq |
N/A | 4.3 ms | N/A |
| 🆕 | Simulation | fsst_like |
N/A | 16.4 ms | N/A |
| 🆕 | Simulation | run_bitpacked_decode |
N/A | 3.1 ms | N/A |
| 🆕 | Simulation | run_bitpacked_eq |
N/A | 1.9 ms | N/A |
| 🆕 | Simulation | run_bitpacked_filter_10pct |
N/A | 4.1 ms | N/A |
| 🆕 | Simulation | run_scan_cost_decode |
N/A | 6.4 ms | N/A |
| 🆕 | Simulation | run_scan_cost_eq |
N/A | 9.5 ms | N/A |
| 🆕 | Simulation | run_scan_cost_filter_10pct |
N/A | 4.9 ms | N/A |
| ... | ... | ... | ... | ... | ... |
ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.
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 codex/cost-model-query-performance (4e73ee6) with ct/compressor-cost-model (43f6c42)
Footnotes
-
9 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. ↩
Codecov Report❌ Patch coverage is ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
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
1 participant
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.
Warning
Experimental: the calibration and PR benchmark feature policy need broader workload validation before this is production-ready.
Summary
ReadTimeCost: estimated read time is encoded bytes divided by effective bandwidth, plus a per-scheme decode charge.ScanCostpreset and use it by default for the stable BtrBlocks scheme pool.SizeCostwhenunstable_encodingsis enabled; the same calibration regressed with that scheme pool.This is stacked on #8753.
Results
TPC-H SF1, DataFusion/Vortex, all 22 queries, two opposite-order runs with 30 iterations each:
Largest improvements were Q4 (-8.86%), Q21 (-4.60%), and Q6 (-3.78%). Q11 regressed by 2.30%.
With unstable encodings enabled, the initial calibration was 1.1% slower in geomean. That is why unstable builds retain
SizeCostand PR benchmarking exercises the stable pool.The focused compression timing sample was blocked by the local Public BI checkout failing its repository update, so this PR does not claim compression-job timing results.