Playing with stream processing on top of Reactive Streams.

Examples in Test.java

Perf test in PerfTest.java

To test performance:

./gradlew benchmarks '-Pjmh=-f 1 -tu s -bm thrpt -wi 5 -i 5 -r 5 -prof GC .*PerfTest.*'

Results: May 22, 2014 - Macbook Pro

Benchmark                                        (size)   Mode   Samples         Mean   Mean error    Units
e.p.PerfTest.liftMultiStep                            1  thrpt         5 15471737.553   197644.846    ops/s
e.p.PerfTest.liftMultiStep                         1024  thrpt         5    12279.060      121.447    ops/s
e.p.PerfTest.liftMultiSubscribe                       1  thrpt         5 14447955.163   287840.378    ops/s
e.p.PerfTest.liftMultiSubscribe                    1024  thrpt         5     8378.420      404.662    ops/s
e.p.PerfTest.liftSingleSubscribe                      1  thrpt         5 26781152.737   383087.870    ops/s
e.p.PerfTest.liftSingleSubscribe                   1024  thrpt         5    16862.970      573.027    ops/s
e.p.PerfTest.processWithSupplier                      1  thrpt         5 27625375.587   585380.799    ops/s
e.p.PerfTest.processWithSupplier                   1024  thrpt         5    16968.807      303.570    ops/s
e.p.PerfTest.processWithSupplierMultiSubscribe        1  thrpt         5 15348067.120   127768.007    ops/s
e.p.PerfTest.processWithSupplierMultiSubscribe     1024  thrpt         5     8529.277      108.034    ops/s
e.p.PerfTest.processWithSupplierMultistep             1  thrpt         5 13377367.313   147047.679    ops/s
e.p.PerfTest.processWithSupplierMultistep          1024  thrpt         5    11976.350       73.694    ops/s
e.p.PerfTest.processWithoutSupplier                   1  thrpt         5 28002030.117   203893.966    ops/s
e.p.PerfTest.processWithoutSupplier                1024  thrpt         5    17249.917      251.941    ops/s

The interesting lines to look at here are:

Benchmark                                        (size)   Mode   Samples         Mean   Mean error    Units
e.p.PerfTest.liftMultiStep                            1  thrpt         5 15471737.553   197644.846    ops/s
e.p.PerfTest.processWithSupplierMultistep             1  thrpt         5 13377367.313   147047.679    ops/s

These are a stream doing 3 transformations, with lift winning at 15.4m ops/second.

e.p.PerfTest.liftSingleSubscribe                      1  thrpt         5 26781152.737   383087.870    ops/s
e.p.PerfTest.processWithSupplier                      1  thrpt         5 27625375.587   585380.799    ops/s

These are a stream doing a single transformation with process winning at 27.6m ops/second.

The analysis of these results is that:

• process is slightly faster (27.6m vs 26.7m ops/second) than lift when doing a single transformation.
• lift is slightly faster (15.4m vs 13.3m ops/second) than process when doing multi-step transformations.