Add Jmh benchmarks along with corresponding rust benchmarks#8597
Open
robert3005 wants to merge 4 commits into
Open
Add Jmh benchmarks along with corresponding rust benchmarks#8597robert3005 wants to merge 4 commits into
robert3005 wants to merge 4 commits into
Conversation
Merging this PR will improve performance by 12.06%
|
| Mode | Benchmark | BASE |
HEAD |
Efficiency | |
|---|---|---|---|---|---|
| ⚡ | Simulation | bitwise_not_vortex_buffer_mut[128] |
244.4 ns | 215.3 ns | +13.55% |
| ⚡ | Simulation | bitwise_not_vortex_buffer_mut[1024] |
304.7 ns | 275.6 ns | +10.58% |
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 mp/jni-bench (b0360cb) with develop (f8b18d9)
Footnotes
-
4 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. ↩
New `vortex-jni-bench` module (JMH) that stresses the vortex-jni read boundary — JNI plus the Arrow C Data Interface — which is the path an Iceberg FormatModel takes to read Vortex from the JVM. Three query shapes (full scan, projection, selective filter) over a synthetic six-column table, consumed column-at-a-time so the numbers reflect format/boundary cost rather than per-row JVM allocation. Includes a batch-granularity diagnostic (Vortex coalesces to ~64K-row read batches regardless of write chunk) and a README with run instructions. Must run against a --release native lib (VORTEX_SKIP_MAKE_TEST_FILES=true to preserve it). v2 TODO: a native Rust criterion read of the same file as a floor, to quote boundary overhead vs native. Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: mprammer <martin@spiraldb.com>
…ix units, add guards A codex-run gauntlet (fresh/correctness/maint) flagged the first cut as overclaiming relative to what it measured. This commit fixes that: - Isolate native pushdown from JVM-side work: add projectionControl (full scan, consume id,y in Java) and filterControl (full scan, filter cat='alpha' in Java). The pushdown speedup is now projection-vs-projectionControl (~4.1x) and selectiveFilter-vs-filterControl (~4.6x), not the confounded ~6x-vs-fullScan. (M2) - fullScan now consumes all six columns at the buffer level (z, cat, tag added), so the "all-six-column scan" number is honest (~40M rows/s). (M1) - @OperationsPerInvocation(ROWS) so JMH reports input rows/s directly, not scans/s. (M3) - @setup validates the file before measuring: exact row count, cat='alpha' returns ROWS/|CATS|, projection schema is exactly [id,y] — fast garbage can't be cited. (M5) - Gradle guard fails the jmh task unless VORTEX_SKIP_MAKE_TEST_FILES=true, so a plain run can't silently rebuild + measure the debug lib. (M6) - @threads(1); tag carries a 10% null rate; README documents the synthetic-data caveats. Read path returns string columns as VarCharVector (Utf8), not ViewVarCharVector — matches the existing TestMinimal read path. Native floor for a boundary-overhead % remains the v2 TODO (M4). Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: mprammer <martin@spiraldb.com>
Adds a native-Rust read "floor" for the vortex-jni read-boundary benchmark so Appendix B can quote JNI+Arrow-C-boundary throughput against a Rust-native floor from one reproducible box. - vortex-jni/src/bin/floor.rs: mirrors the JNI read path (session -> file -> scan -> array stream -> execute_arrow) on the same CurrentThreadRuntime the JNI lib uses, stopping before the Arrow C export / JNI / JVM. Reports rows/s for fullScan / projection(id,y) / selectiveFilter(cat='alpha'), normalized by the file row count to match JMH's @OperationsPerInvocation. Validates row counts. - VortexJniReadBenchmark: when VORTEX_JNI_BENCH_FILE is set, write+keep the table at that path (instead of a deleted temp file) so the floor reads the byte-identical file. Default behavior unchanged. Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: mprammer <martin@spiraldb.com>
…ruption underflow.rs reproduces whole-row data corruption in the vortex-jni Arrow-C write path: a multi-chunk write of Arrow-imported variable-width (Utf8/Utf8View) columns with a reused VectorSchemaRoot and an uneven final chunk corrupts the tail of full-size chunks (strings -> "", ints -> 0). A pure-Rust write of the same data is clean; the JMH read bench's `cat = 'alpha'` filter consequently under-counts (111,168 vs the expected 125,000 = ROWS/16). Run: cargo run -p vortex-jni --bin underflow (knobs: UF_ROWS / UF_CHUNK / UF_VIEW) Co-Authored-By: Claude <noreply@anthropic.com> Signed-off-by: mprammer <martin@spiraldb.com>
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.
Add Jmh benchmarks for java bindings with corresponding rust version of those
benchmarks