Open
Conversation
connortsui20
added
action/benchmark
github-actions
bot
removed
the
action/benchmark
This was referenced Apr 1, 2026
Contributor
Polar Signals Profiling ResultsLatest Run
Previous Runs (7)
Powered by Polar Signals Cloud |
Contributor
Benchmarks: PolarSignals ProfilingVortex (geomean): 1.008x ➖ datafusion / vortex-file-compressed (1.008x ➖, 0↑ 0↓)
|
Contributor
Benchmarks: TPC-H SF=10 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (1.027x ➖, 0↑ 0↓)
datafusion / vortex-compact (1.019x ➖, 0↑ 0↓)
datafusion / parquet (1.028x ➖, 0↑ 1↓)
datafusion / arrow (1.093x ➖, 0↑ 11↓)
duckdb / vortex-file-compressed (1.030x ➖, 0↑ 1↓)
duckdb / vortex-compact (1.035x ➖, 0↑ 0↓)
duckdb / parquet (1.018x ➖, 0↑ 1↓)
duckdb / duckdb (1.046x ➖, 0↑ 1↓)
Full attributed analysis
|
Contributor
Benchmarks: TPC-H SF=1 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (0.893x ➖, 5↑ 0↓)
datafusion / vortex-compact (1.009x ➖, 1↑ 1↓)
datafusion / parquet (1.133x ➖, 0↑ 7↓)
duckdb / vortex-file-compressed (1.011x ➖, 1↑ 0↓)
duckdb / vortex-compact (1.064x ➖, 0↑ 1↓)
duckdb / parquet (1.051x ➖, 0↑ 2↓)
Full attributed analysis
|
Contributor
Benchmarks: FineWeb S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (0.804x ➖, 2↑ 0↓)
datafusion / vortex-compact (0.824x ➖, 2↑ 0↓)
datafusion / parquet (0.830x ➖, 2↑ 0↓)
duckdb / vortex-file-compressed (0.947x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.985x ➖, 0↑ 0↓)
duckdb / parquet (1.020x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: Clickbench on NVMEVerdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (0.941x ➖, 15↑ 0↓)
datafusion / parquet (0.952x ➖, 12↑ 1↓)
duckdb / vortex-file-compressed (0.929x ➖, 14↑ 2↓)
duckdb / parquet (0.970x ➖, 0↑ 0↓)
duckdb / duckdb (0.944x ➖, 5↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: Statistical and Population GeneticsVerdict: No clear signal (low confidence) duckdb / vortex-file-compressed (0.950x ➖, 1↑ 0↓)
duckdb / vortex-compact (1.032x ➖, 0↑ 1↓)
duckdb / parquet (0.964x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: TPC-H SF=1 on NVMEVerdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (0.994x ➖, 0↑ 0↓)
datafusion / vortex-compact (0.990x ➖, 0↑ 0↓)
datafusion / parquet (1.011x ➖, 1↑ 3↓)
datafusion / arrow (1.006x ➖, 0↑ 1↓)
duckdb / vortex-file-compressed (1.018x ➖, 0↑ 0↓)
duckdb / vortex-compact (1.007x ➖, 0↑ 0↓)
duckdb / parquet (1.000x ➖, 2↑ 2↓)
duckdb / duckdb (1.006x ➖, 0↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: FineWeb NVMeVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.860x ✅, 4↑ 0↓)
datafusion / vortex-compact (0.935x ➖, 0↑ 0↓)
datafusion / parquet (0.901x ➖, 4↑ 0↓)
duckdb / vortex-file-compressed (0.915x ➖, 3↑ 0↓)
duckdb / vortex-compact (0.900x ✅, 3↑ 0↓)
duckdb / parquet (0.921x ➖, 2↑ 0↓)
Full attributed analysis
|
Contributor
Benchmarks: TPC-DS SF=1 on NVMEVerdict: No clear signal (low confidence) datafusion / vortex-file-compressed (0.843x ✅, 94↑ 0↓)
datafusion / vortex-compact (0.908x ➖, 46↑ 2↓)
datafusion / parquet (0.898x ✅, 50↑ 0↓)
duckdb / vortex-file-compressed (1.023x ➖, 9↑ 16↓)
duckdb / vortex-compact (1.084x ➖, 3↑ 36↓)
duckdb / parquet (1.011x ➖, 0↑ 2↓)
duckdb / duckdb (1.016x ➖, 1↑ 1↓)
Full attributed analysis
|
Contributor
Benchmarks: Random AccessVortex (geomean): 0.946x ➖ unknown / unknown (1.032x ➖, 7↑ 21↓)
|
Contributor
Benchmarks: TPC-H SF=10 on S3Verdict: No clear signal (environment too noisy confidence) datafusion / vortex-file-compressed (0.909x ➖, 2↑ 1↓)
datafusion / vortex-compact (0.783x ➖, 7↑ 0↓)
datafusion / parquet (0.627x ✅, 15↑ 0↓)
duckdb / vortex-file-compressed (0.941x ➖, 0↑ 0↓)
duckdb / vortex-compact (0.969x ➖, 0↑ 0↓)
duckdb / parquet (0.928x ➖, 0↑ 1↓)
Full attributed analysis
|
Contributor
Benchmarks: CompressionVortex (geomean): 1.000x ➖ unknown / unknown (0.987x ➖, 11↑ 9↓)
|
connortsui20
force-pushed
the
ct/compress-estimation
branch
from
32cd06a to
45acd04
Compare
connortsui20
added
action/benchmark
connortsui20
force-pushed
the
ct/compress-estimation
branch
3 times, most recently
from
6887c12 to
b8ec852
Compare
connortsui20
removed
the
action/benchmark
connortsui20
force-pushed
the
ct/compress-estimation
branch
2 times, most recently
from
a7be89a to
7c6a0f5
Compare
github-actions
bot
removed
the
action/benchmark
connortsui20
commented
Apr 1, 2026
Comment on lines
+368
to
+383
| let estimate = estimate_callback(self, data, ctx.clone())?; | ||
|
|
||
| match estimate { | ||
| CompressionEstimate::Skip => {} | ||
| CompressionEstimate::AlwaysUse => return Ok(Some(scheme)), | ||
| CompressionEstimate::Ratio(ratio) => { | ||
| if is_better_ratio(ratio, &best) { | ||
| best = Some((scheme, ratio)); | ||
| } | ||
| } | ||
| e @ (CompressionEstimate::Sample | CompressionEstimate::Estimate(_)) => { | ||
| vortex_panic!( | ||
| "an estimation function returned an invalid variant {e:?}" | ||
| ) | ||
| } | ||
| } |
Contributor
Author
There was a problem hiding this comment.
We probably want the estimate_callback to return something else other than CompressionEstimate, maybe call it ResolvedEstimate.
a10y
reviewed
Apr 1, 2026
a10y
reviewed
Apr 1, 2026
Contributor
|
Might wanna double-check we're not making any of the TPC-H files bigger by accident with this change? Though this could be noise 
|
a10y
reviewed
Apr 1, 2026
Contributor
|
Maybe in a follow up or maybe before, shall we add to the size of all files for all sql benchmarks. |
joseph-isaacs
added
the
action/benchmark
github-actions
bot
removed
the
action/benchmark
Contributor
|
Did you expect to a load of size wins |
Contributor
|
Contributor
Author
|
To be honest I have no idea what the compressor is actually choosing, I just changed the logic in a way that makes sense to me rather than figuring out what is best for performance. |
connortsui20
force-pushed
the
ct/compress-estimation
branch
from
5ad9c52 to
7931da7
Compare
Contributor
|
That comment doesn't inspire confidence in this change |
Contributor
|
I think you should look at diff of file sizes for all benchmarks before we merge this? |
Contributor
Author
|
Well that's part of the reason for this change, we want to move towards having better introspection to the compressor by making its behavior more understandable. I'm happy to go all the way (smarter logic for what schemes we try first, better I'm going to bench things locally rather than rely on the CI here because I frankly do not trust any of the numbers (other than the compression size which is the same or better). |
github-actions
bot
removed
the
action/benchmark
Contributor
|
I think we should add file size measures for all sql benchmarks |
Contributor
Author
|
I really just don't trust our SQL benchmarks on PRs at all. I've ran it several times here and every time it is a completely different result. This past time there were more regressions, but that happened for Parquet as well. I think these are just way too noisy to be useful unless EVERYTHING clearly regresses across all benchmarks. To validate, I ran everything locally on my x86 linux machine, and I only saw improvements in file size. I am currently running the runtime benchmarks comparison now. This file size and array tree report is generated by claude but I am pretty confident it is correct because I ran the commands myself. DetailsTPCH SF10 File Size Comparison:
|
| File | Branch (bytes) | Develop (bytes) | Delta | % Change |
|---|---|---|---|---|
| customer_0 | 109,632,916 | 109,638,340 | -5,424 | -0.005% |
| lineitem_0 | 141,498,968 | 141,501,928 | -2,960 | -0.002% |
| lineitem_1 | 141,175,136 | 141,178,464 | -3,328 | -0.002% |
| lineitem_2 | 141,236,712 | 141,239,568 | -2,856 | -0.002% |
| lineitem_3 | 140,773,056 | 140,777,008 | -3,952 | -0.003% |
| lineitem_4 | 141,293,360 | 141,296,480 | -3,120 | -0.002% |
| lineitem_5 | 140,903,880 | 140,907,832 | -3,952 | -0.003% |
| lineitem_6 | 140,373,752 | 140,378,328 | -4,576 | -0.003% |
| lineitem_7 | 140,951,792 | 140,955,328 | -3,536 | -0.003% |
| lineitem_8 | 140,996,280 | 140,999,816 | -3,536 | -0.003% |
| lineitem_9 | 141,439,592 | 141,442,296 | -2,704 | -0.002% |
| lineitem_10 | 141,383,744 | 141,386,864 | -3,120 | -0.002% |
| lineitem_11 | 141,610,176 | 141,612,672 | -2,496 | -0.002% |
| lineitem_12 | 141,390,000 | 141,393,120 | -3,120 | -0.002% |
| nation_0 | 8,032 | 8,704 | -672 | -7.72% |
| orders_0 | 171,523,868 | 171,697,428 | -173,560 | -0.10% |
| orders_1 | 171,752,756 | 171,800,308 | -47,552 | -0.03% |
| orders_2 | 171,587,564 | 171,711,900 | -124,336 | -0.07% |
| part_0 | 28,270,320 | 28,274,224 | -3,904 | -0.01% |
| part_1 | 28,235,864 | 28,239,768 | -3,904 | -0.01% |
| partsupp_0 | 188,516,296 | 188,516,336 | -40 | -0.00% |
| partsupp_1 | 188,699,112 | 188,699,136 | -24 | -0.00% |
| region_0 | 4,844 | 5,260 | -416 | -7.91% |
| supplier_0 | 6,995,400 | 6,995,784 | -384 | -0.01% |
| Total | 2,900,253,420 | 2,900,656,892 | -403,472 | -0.014% |
Branch is 403 KB smaller overall (-0.014%). Every file is equal or smaller.
Compact Format (vortex-compact)
| File | Branch (bytes) | Develop (bytes) | Delta | % Change |
|---|---|---|---|---|
| customer_0 | 77,604,740 | 77,604,740 | 0 | 0.000% |
| lineitem_0 | 105,267,688 | 105,267,688 | 0 | 0.000% |
| lineitem_1 | 105,324,192 | 105,318,688 | +5,504 | +0.005% |
| lineitem_2 | 105,264,840 | 105,259,336 | +5,504 | +0.005% |
| lineitem_3 | 105,295,552 | 105,290,048 | +5,504 | +0.005% |
| lineitem_4 | 105,227,424 | 105,227,424 | 0 | 0.000% |
| lineitem_5 | 105,376,392 | 105,376,392 | 0 | 0.000% |
| lineitem_6 | 105,372,336 | 105,355,832 | +16,504 | +0.016% |
| lineitem_7 | 105,260,576 | 105,255,080 | +5,496 | +0.005% |
| lineitem_8 | 105,276,392 | 105,265,384 | +11,008 | +0.010% |
| lineitem_9 | 105,131,256 | 105,131,256 | 0 | 0.000% |
| lineitem_10 | 105,299,184 | 105,293,680 | +5,504 | +0.005% |
| lineitem_11 | 105,200,696 | 105,195,184 | +5,512 | +0.005% |
| lineitem_12 | 105,303,320 | 105,297,808 | +5,512 | +0.005% |
| nation_0 | 6,168 | 6,840 | -672 | -9.82% |
| orders_0 | 120,127,588 | 120,127,588 | 0 | 0.000% |
| orders_1 | 120,096,156 | 120,096,156 | 0 | 0.000% |
| orders_2 | 120,117,156 | 120,117,156 | 0 | 0.000% |
| part_0 | 18,934,712 | 18,934,712 | 0 | 0.000% |
| part_1 | 18,935,656 | 18,935,656 | 0 | 0.000% |
| partsupp_0 | 132,812,400 | 132,812,400 | 0 | 0.000% |
| partsupp_1 | 132,790,680 | 132,790,680 | 0 | 0.000% |
| region_0 | 4,340 | 4,756 | -416 | -8.75% |
| supplier_0 | 4,945,512 | 4,945,512 | 0 | 0.000% |
| Total | 2,114,974,956 | 2,114,909,996 | +64,960 | +0.003% |
Branch is 65 KB larger overall (+0.003%). Effectively noise.
Encoding Differences (Array Tree Analysis)
For files with layout tree differences, we materialized the full array encoding trees and diffed them. Three patterns explain all observed changes:
1. FoR eliminates bitpack patches (lineitem l_orderkey run-ends, all default partitions)
Branch wraps bitpacked run-end offsets in Frame-of-Reference, which shifts the value range down so all values fit without patches:
Branch: ends: fastlanes.for(u32) -> fastlanes.bitpacked(u32, bit_width=17, patches: None)
Develop: ends: fastlanes.bitpacked(u32, bit_width=17, patches: Some(...))
+ patch_indices (2 B) + patch_values (4 B) + patch_chunk_offsets (32 B)
FoR subtracts base=6, so max 131072 becomes 131066 which fits in 17 bits. Without FoR, 131072 requires a patch. Net savings: ~38 bytes/chunk metadata overhead for patches, repeated across every chunk in every lineitem partition.
2. FSST chosen over dict+FSST for o_clerk (orders, some chunks)
Branch: o_clerk: vortex.fsst(utf8, len=8192) -- 33.88 kB
Develop: o_clerk: vortex.dict(utf8, len=8192) -- 36.65 kB
codes: fastlanes.bitpacked(u16, bit_width=13) -- 13.31 kB
values: vortex.fsst(utf8, len=5562) -- 23.34 kB
Branch correctly identifies that FSST alone (33.88 kB) beats dict+FSST (36.65 kB) for the o_clerk column which has 5562 distinct values across 8192 rows (68% unique). This is the primary driver of the ~170 KB/file savings on orders.
3. FoR wrapping on o_orderkey (orders, all chunks)
Branch: o_orderkey: fastlanes.for(i64) -> fastlanes.bitpacked(i64, bit_width=19)
Develop: o_orderkey: fastlanes.bitpacked(i64, bit_width=19)
Same buffer size (19.46 kB) -- FoR adds a small metadata scalar but no extra buffer. Neutral to slightly beneficial.
4. Dict vs PCO for l_quantity MSP (compact lineitem, rare chunks)
Branch: msp: vortex.dict(i16, len=8192) -> codes: bitpacked(u8) + values: pco(i16) -- 6.17 kB
Develop: msp: vortex.pco(i16, len=8192) -- 92.79 kB
Branch picks dictionary encoding for the decimal quantity MSP which is dramatically better for this specific chunk (15x). However, this only occurs in a few isolated chunks within the compact format, and the net file-level effect is negligible.
Summary
- Default format: Branch is 403 KB smaller (-0.014%). No regressions.
- Compact format: Branch is 65 KB larger (+0.003%). Effectively identical.
- All encoding changes are explainable and reasonable. The estimation rework produces equal or better compression decisions on TPCH SF10.
Contributor
|
what happened to |
Contributor
Author
|
Edit: Whoops I forgot to rebase on develop so the runtime results are not reliable |
Signed-off-by: Connor Tsui <connor.tsui20@gmail.com>
connortsui20
force-pushed
the
ct/compress-estimation
branch
from
7931da7 to
6821f7f
Compare
github-actions
bot
removed
the
action/benchmark
Contributor
Author
|
Ok yeah there are effectively no regressions. I ran the both versions of files against both my branch and develop and there is no difference. I think the earlier regressions was from the regression we saw that was recently fixed. DetailsTPCH SF10 Runtime Comparison:
|
| Query | Branch (ms) | Develop (ms) | Delta (ms) | % Change |
|---|---|---|---|---|
| tpch_q01 | 466.7 | 468.5 | -1.8 | -0.4% |
| tpch_q02 | 152.8 | 152.4 | +0.3 | +0.2% |
| tpch_q03 | 318.6 | 320.8 | -2.1 | -0.7% |
| tpch_q04 | 121.5 | 124.0 | -2.5 | -2.0% |
| tpch_q05 | 635.9 | 630.2 | +5.7 | +0.9% |
| tpch_q06 | 62.3 | 61.7 | +0.6 | +1.0% |
| tpch_q07 | 893.6 | 904.1 | -10.5 | -1.2% |
| tpch_q08 | 620.3 | 613.6 | +6.7 | +1.1% |
| tpch_q09 | 1145.8 | 1137.2 | +8.6 | +0.8% |
| tpch_q10 | 360.3 | 359.1 | +1.2 | +0.3% |
| tpch_q11 | 124.9 | 125.5 | -0.6 | -0.5% |
| tpch_q12 | 148.8 | 150.6 | -1.7 | -1.1% |
| tpch_q13 | 201.0 | 203.4 | -2.4 | -1.2% |
| tpch_q14 | 83.9 | 84.5 | -0.6 | -0.7% |
| tpch_q15 | 158.9 | 155.8 | +3.1 | +2.0% |
| tpch_q16 | 101.0 | 100.2 | +0.8 | +0.8% |
| tpch_q17 | 982.8 | 981.3 | +1.5 | +0.2% |
| tpch_q18 | 1300.9 | 1299.8 | +1.1 | +0.1% |
| tpch_q19 | 87.6 | 88.0 | -0.4 | -0.5% |
| tpch_q20 | 261.4 | 258.3 | +3.1 | +1.2% |
| tpch_q21 | 974.0 | 989.0 | -15.0 | -1.5% |
| tpch_q22 | 61.8 | 60.8 | +1.0 | +1.6% |
End-to-end — Compact (vortex-compact)
| Query | Branch (ms) | Develop (ms) | Delta (ms) | % Change |
|---|---|---|---|---|
| tpch_q01 | 462.4 | 466.8 | -4.4 | -1.0% |
| tpch_q02 | 157.5 | 156.9 | +0.7 | +0.4% |
| tpch_q03 | 309.3 | 311.4 | -2.1 | -0.7% |
| tpch_q04 | 122.4 | 122.9 | -0.6 | -0.5% |
| tpch_q05 | 624.5 | 623.0 | +1.5 | +0.2% |
| tpch_q06 | 65.5 | 63.0 | +2.4 | +3.9% |
| tpch_q07 | 886.2 | 894.6 | -8.4 | -0.9% |
| tpch_q08 | 610.4 | 608.3 | +2.1 | +0.3% |
| tpch_q09 | 1143.7 | 1128.1 | +15.6 | +1.4% |
| tpch_q10 | 361.8 | 367.4 | -5.6 | -1.5% |
| tpch_q11 | 129.3 | 128.2 | +1.0 | +0.8% |
| tpch_q12 | 173.4 | 173.0 | +0.4 | +0.2% |
| tpch_q13 | 213.6 | 215.4 | -1.8 | -0.8% |
| tpch_q14 | 89.0 | 91.2 | -2.3 | -2.5% |
| tpch_q15 | 191.0 | 188.2 | +2.8 | +1.5% |
| tpch_q16 | 106.4 | 107.9 | -1.5 | -1.4% |
| tpch_q17 | 973.4 | 977.7 | -4.4 | -0.4% |
| tpch_q18 | 1271.3 | 1269.5 | +1.8 | +0.1% |
| tpch_q19 | 111.2 | 110.7 | +0.6 | +0.5% |
| tpch_q20 | 258.5 | 262.2 | -3.7 | -1.4% |
| tpch_q21 | 951.9 | 966.8 | -14.9 | -1.5% |
| tpch_q22 | 64.8 | 67.5 | -2.8 | -4.1% |
2. File differences only: branch files vs develop files (same develop binary)
Same binary (develop) — Default (vortex-file-compressed)
| Query | Branch files (ms) | Develop files (ms) | Delta (ms) | % Change |
|---|---|---|---|---|
| tpch_q01 | 464.0 | 468.5 | -4.5 | -1.0% |
| tpch_q02 | 151.8 | 152.4 | -0.6 | -0.4% |
| tpch_q03 | 320.4 | 320.8 | -0.3 | -0.1% |
| tpch_q04 | 126.5 | 124.0 | +2.5 | +2.0% |
| tpch_q05 | 638.7 | 630.2 | +8.5 | +1.4% |
| tpch_q06 | 61.7 | 61.7 | +0.0 | +0.0% |
| tpch_q07 | 898.8 | 904.1 | -5.3 | -0.6% |
| tpch_q08 | 610.2 | 613.6 | -3.4 | -0.5% |
| tpch_q09 | 1138.8 | 1137.2 | +1.6 | +0.1% |
| tpch_q10 | 357.6 | 359.1 | -1.4 | -0.4% |
| tpch_q11 | 124.2 | 125.5 | -1.3 | -1.1% |
| tpch_q12 | 152.3 | 150.6 | +1.8 | +1.2% |
| tpch_q13 | 201.7 | 203.4 | -1.7 | -0.8% |
| tpch_q14 | 85.5 | 84.5 | +1.0 | +1.1% |
| tpch_q15 | 157.4 | 155.8 | +1.6 | +1.0% |
| tpch_q16 | 103.3 | 100.2 | +3.1 | +3.1% |
| tpch_q17 | 984.6 | 981.3 | +3.3 | +0.3% |
| tpch_q18 | 1298.1 | 1299.8 | -1.7 | -0.1% |
| tpch_q19 | 87.3 | 88.0 | -0.6 | -0.7% |
| tpch_q20 | 257.6 | 258.3 | -0.7 | -0.3% |
| tpch_q21 | 976.3 | 989.0 | -12.7 | -1.3% |
| tpch_q22 | 60.2 | 60.8 | -0.6 | -1.0% |
Same binary (develop) — Compact (vortex-compact)
| Query | Branch files (ms) | Develop files (ms) | Delta (ms) | % Change |
|---|---|---|---|---|
| tpch_q01 | 468.8 | 466.8 | +2.0 | +0.4% |
| tpch_q02 | 156.8 | 156.9 | -0.1 | -0.1% |
| tpch_q03 | 308.2 | 311.4 | -3.1 | -1.0% |
| tpch_q04 | 121.8 | 122.9 | -1.1 | -0.9% |
| tpch_q05 | 623.8 | 623.0 | +0.8 | +0.1% |
| tpch_q06 | 63.2 | 63.0 | +0.1 | +0.2% |
| tpch_q07 | 889.0 | 894.6 | -5.6 | -0.6% |
| tpch_q08 | 610.8 | 608.3 | +2.5 | +0.4% |
| tpch_q09 | 1123.5 | 1128.1 | -4.6 | -0.4% |
| tpch_q10 | 363.6 | 367.4 | -3.8 | -1.0% |
| tpch_q11 | 128.5 | 128.2 | +0.3 | +0.2% |
| tpch_q12 | 170.1 | 173.0 | -2.9 | -1.7% |
| tpch_q13 | 214.1 | 215.4 | -1.3 | -0.6% |
| tpch_q14 | 89.3 | 91.2 | -1.9 | -2.1% |
| tpch_q15 | 187.9 | 188.2 | -0.3 | -0.1% |
| tpch_q16 | 109.1 | 107.9 | +1.2 | +1.1% |
| tpch_q17 | 975.0 | 977.7 | -2.7 | -0.3% |
| tpch_q18 | 1278.1 | 1269.5 | +8.6 | +0.7% |
| tpch_q19 | 110.4 | 110.7 | -0.3 | -0.2% |
| tpch_q20 | 260.5 | 262.2 | -1.7 | -0.6% |
| tpch_q21 | 976.9 | 966.8 | +10.1 | +1.0% |
| tpch_q22 | 67.1 | 67.5 | -0.5 | -0.7% |
3. Binary differences only: branch binary vs develop binary (same develop files)
Same files (develop) — Default (vortex-file-compressed)
| Query | Branch binary (ms) | Develop binary (ms) | Delta (ms) | % Change |
|---|---|---|---|---|
| tpch_q01 | 477.4 | 468.5 | +8.9 | +1.9% |
| tpch_q02 | 151.5 | 152.4 | -0.9 | -0.6% |
| tpch_q03 | 320.3 | 320.8 | -0.5 | -0.1% |
| tpch_q04 | 123.0 | 124.0 | -1.0 | -0.8% |
| tpch_q05 | 643.1 | 630.2 | +12.9 | +2.0% |
| tpch_q06 | 63.7 | 61.7 | +2.0 | +3.3% |
| tpch_q07 | 896.4 | 904.1 | -7.7 | -0.8% |
| tpch_q08 | 614.6 | 613.6 | +1.0 | +0.2% |
| tpch_q09 | 1138.3 | 1137.2 | +1.1 | +0.1% |
| tpch_q10 | 357.7 | 359.1 | -1.3 | -0.4% |
| tpch_q11 | 125.4 | 125.5 | -0.2 | -0.1% |
| tpch_q12 | 148.7 | 150.6 | -1.9 | -1.3% |
| tpch_q13 | 202.1 | 203.4 | -1.3 | -0.6% |
| tpch_q14 | 84.7 | 84.5 | +0.2 | +0.3% |
| tpch_q15 | 157.9 | 155.8 | +2.1 | +1.3% |
| tpch_q16 | 101.0 | 100.2 | +0.8 | +0.8% |
| tpch_q17 | 981.4 | 981.3 | +0.2 | +0.0% |
| tpch_q18 | 1308.1 | 1299.8 | +8.3 | +0.6% |
| tpch_q19 | 86.2 | 88.0 | -1.7 | -2.0% |
| tpch_q20 | 258.8 | 258.3 | +0.5 | +0.2% |
| tpch_q21 | 971.6 | 989.0 | -17.4 | -1.8% |
| tpch_q22 | 60.9 | 60.8 | +0.2 | +0.3% |
Same files (develop) — Compact (vortex-compact)
| Query | Branch binary (ms) | Develop binary (ms) | Delta (ms) | % Change |
|---|---|---|---|---|
| tpch_q01 | 466.9 | 466.8 | +0.1 | +0.0% |
| tpch_q02 | 157.4 | 156.9 | +0.6 | +0.4% |
| tpch_q03 | 311.3 | 311.4 | -0.1 | -0.0% |
| tpch_q04 | 124.5 | 122.9 | +1.6 | +1.3% |
| tpch_q05 | 628.5 | 623.0 | +5.5 | +0.9% |
| tpch_q06 | 65.2 | 63.0 | +2.2 | +3.5% |
| tpch_q07 | 894.3 | 894.6 | -0.3 | -0.0% |
| tpch_q08 | 607.5 | 608.3 | -0.9 | -0.1% |
| tpch_q09 | 1125.7 | 1128.1 | -2.4 | -0.2% |
| tpch_q10 | 362.3 | 367.4 | -5.1 | -1.4% |
| tpch_q11 | 126.9 | 128.2 | -1.3 | -1.0% |
| tpch_q12 | 174.9 | 173.0 | +1.9 | +1.1% |
| tpch_q13 | 210.7 | 215.4 | -4.6 | -2.1% |
| tpch_q14 | 90.4 | 91.2 | -0.8 | -0.8% |
| tpch_q15 | 190.3 | 188.2 | +2.1 | +1.1% |
| tpch_q16 | 108.8 | 107.9 | +0.9 | +0.8% |
| tpch_q17 | 978.4 | 977.7 | +0.6 | +0.1% |
| tpch_q18 | 1266.4 | 1269.5 | -3.1 | -0.2% |
| tpch_q19 | 110.5 | 110.7 | -0.2 | -0.2% |
| tpch_q20 | 261.5 | 262.2 | -0.7 | -0.3% |
| tpch_q21 | 955.5 | 966.8 | -11.2 | -1.2% |
| tpch_q22 | 64.7 | 67.5 | -2.9 | -4.3% |
Contributor
|
Thanks for confirming, I will have a look tomorrow |
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
4 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
Tracking Issue: #7216
Adds a new
CompressionEstimatetype invortex-compressor/src/estimate.rsthat theexpected_compression_ratiomethod now returns. Additionally moves some things around for clarity.Note that this is not just a refactor, there is subtle logic that has changed in a few places (that I think is better, not actually sure). I'm happy to split out some stuff into other PRs if that helps.
Future Work
Exactthat returns the fully compressed array in the case that we can only determine if a scheme is a candidate by compressing the whole thing without any errors; the only case where we want to do this isSequenceArray(and maybe there's an argument to do this forConstantArraytoo, but the semantics aroundConstantArrayshould be even more special regardless, imo).ResolvedEstimateenum instead.ConstantSchemelogic into the compressor since I cannot think of any reason why you would not want to have aConstantScheme(except when you have a very small array, and at that point you don't care about perf regardless).API Changes
expected_compression_rationow only takes stats and compressor context (it does not take the compressor at all) and returns aCompressionEstimate. This method must be super quick, and any sampling or expensive operations are now deferred to later by the compressor.Testing
Just a few extra tests, am relying on the existing test suite as it's not like completely new logic is happening.