Hi,
I've created a following JIRA task, to add JMH dependency and allow
execution of microbenchmarks based on JMH framework. I've added maven
profile to separate compilation of benchmark related jar from the main
stream. My proposed changes attached to the JIRA issue, any comments or
suggestions are welcomed.
Also it will be interesting to hear ideas where to actually write and
implement the benchmarks once such change will be added. Moreover it's
important to note, that using JMH it only allows to compare algorithms and
different implementations performance wise rather than comparing the
actually accuracy of result produced. For example if I'll add a benchmark
to compare different random source providers, I will receive results which
show me what is the most performant provider for given workload, while
nothing regarding the actually accuracy of the algorithm I've implemented
to run the comparison.
More particularly I've added a benchmark to estimate value of PI add this
is the results I've got out of JMH:
# Run complete. Total time: 00:01:35
Benchmark (pairsToGenerate) (randomSourceName)
Mode Cnt Score Error Units
PiComputationBenchmark.computePi 1000000 JDK
avgt 5 44076.310 ± 2794.888 us/op
PiComputationBenchmark.computePi 1000000 WELL_512_A
avgt 5 19374.112 ± 4973.542 us/op
PiComputationBenchmark.computePi 1000000 WELL_1024_A
avgt 5 20575.240 ± 4298.444 us/op
PiComputationBenchmark.computePi 1000000 WELL_19937_A
avgt 5 42208.136 ± 2062.648 us/op
PiComputationBenchmark.computePi 1000000 WELL_19937_C
avgt 5 45105.231 ± 2752.918 us/op
PiComputationBenchmark.computePi 1000000 WELL_44497_A
avgt 5 49710.663 ± 15786.830 us/op
PiComputationBenchmark.computePi 1000000 WELL_44497_B
avgt 5 48984.700 ± 2449.719 us/op
PiComputationBenchmark.computePi 1000000 MT
avgt 5 22466.565 ± 1377.585 us/op
PiComputationBenchmark.computePi 1000000 ISAAC
avgt 5 21615.283 ± 1080.331 us/op
Note, there is nothing here that shows how actually computed PI results is
close to real value.
Best regards,
Artem Barger.
