Describe the enhancement requested
similar to what I'm doing for iceberg in apache/iceberg#15628 , add a jmh benchmark to measure variant performance
- build
- serialize
- deserialize
- shed
Some of these operations take microseconds, and while x86 has the RDTSC opcode for nS granularity, ARM parts don't seem as accurate, there's a huge variance here. All benchmarks will be designed to take 10s or 100s of uS though multiple iterations, with the slower operations scaled by the same amount to make comparision easier, especially when looking at graphs
Full written up results
(https://steveloughran.github.io/benchmarking-variants/)[Benchmarking Parquet Variants through Iceberg]
Component(s)
- parquet-benchmarks
- parquet-variant (one initial benchhmark driven change)
Describe the enhancement requested
similar to what I'm doing for iceberg in apache/iceberg#15628 , add a jmh benchmark to measure variant performance
Some of these operations take microseconds, and while x86 has the
RDTSCopcode for nS granularity, ARM parts don't seem as accurate, there's a huge variance here. All benchmarks will be designed to take 10s or 100s of uS though multiple iterations, with the slower operations scaled by the same amount to make comparision easier, especially when looking at graphsFull written up results
(https://steveloughran.github.io/benchmarking-variants/)[Benchmarking Parquet Variants through Iceberg]
Component(s)