richardstartin opened a new pull request #8065: URL: https://github.com/apache/pinot/pull/8065
This adds a benchmark which measures FixedBitSVReader performance when there are multiple cardinalities and therefore multiple implementations present, which illustrates that existing benchmarks are unrealistically kind. For example, when there are 3 or more distinct cardinalities, performance suffers because reader calls no longer inline: ``` Benchmark (_numBits0) (_numBits1) (_numBits2) Mode Cnt Score Error Units BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.dense 1 1 1 avgt 5 70.743 ± 0.442 us/op BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.sequential 1 1 1 avgt 5 12.764 ± 0.285 us/op BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.sparse 1 1 1 avgt 5 70.897 ± 0.514 us/op Benchmark (_numBits0) (_numBits1) (_numBits2) Mode Cnt Score Error Units BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.dense 2 2 2 avgt 5 67.808 ± 2.751 us/op BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.sequential 2 2 2 avgt 5 12.931 ± 0.187 us/op BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.sparse 2 2 2 avgt 5 75.187 ± 0.160 us/op Benchmark (_numBits0) (_numBits1) (_numBits2) Mode Cnt Score Error Units BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.dense 1 2 2 avgt 5 73.471 ± 0.607 us/op BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.sequential 1 2 2 avgt 5 13.156 ± 0.513 us/op BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.sparse 1 2 2 avgt 5 78.023 ± 1.484 us/op Benchmark (_numBits0) (_numBits1) (_numBits2) Mode Cnt Score Error Units BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.dense 1 2 3 avgt 5 106.137 ± 1.831 us/op BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.sequential 1 2 3 avgt 5 13.132 ± 0.259 us/op BenchmarkFixedBitSVForwardIndexReaderMixedCardinalities.sparse 1 2 3 avgt 5 106.371 ± 3.274 us/op ``` -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
