commons-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alex D Herbert (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (RNG-72) Create a RandomSource.create benchmark
Date Fri, 01 Mar 2019 11:47:00 GMT

    [ https://issues.apache.org/jira/browse/RNG-72?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16781598#comment-16781598
] 

Alex D Herbert commented on RNG-72:
-----------------------------------

I have compiled a new listing using the output of {{ConstructionPerformance}} and {{GenerationPerformance}}.

For reference the construction benchmark creates 500 RNGs, using pre-built seeds where appropriate.
Test the following construction methods:
 * Create using the native constructor with the *new* keyword
 * Create using a cached *{{Constructor<Object>}}* with *{{newInstance}}*
 * Create using a cached *{{Class<?>}}*, lookup the *{{Constructor<Object>}}*
and use *{{newInstance}}*
 * Create using *{{RandomSource}}* with the *native* seed
 * Create using *{{RandomSource}}* with a *{{null}}* seed
 * Create using *{{RandomSource}}* with a *{{Long}}* seed
 * Create using *{{RandomSource}}* with a *truncated native* seed forcing self seeding
 * Create using *{{RandomSource}}* with a *{{byte[]}}* of the appropriate length

The time for construction of a single generator is the benchmark time divided by 500.

Generation time is shown for the given number of values. For simplicity this is created by
linearly scaling the generation time for 1,000,000 values. The time uses the native method
of the generator (either {{nextInt}} or {{nextLong}}) so *is not comparable between generators
of different types*. It is indicative of the speed it will take to create a generator and
output values.

The table is constructed by adding the time for construction to the generation time. Only
the total time is shown for the construction using *new*. All other times are relative to
this. The table is sorted by speed of the *new* construction and grouped by the number of
values generated.

*Note: This listing involves adding of times from two benchmarks run on the same computer,
using the same JMH benchmark settings for average time in microseconds. An independent test
using construction and generation together indicated this is a valid approach as the numbers
for creating 1,000,000 values were within 10%.*
||Values||Generation||RandomSource||new||newInstance||lookupNewInstance||createNativeSeed||createNullSeed||createLongSeed||createSelfSeed||createByteArray||
|0|0|SPLIT_MIX_64|0.0091|1.89|5.63|8.62|13.48|9.11|0|12.11|
|0|0|KISS|0.02132|1.37|2.82|4.07|67.38|10.79|4.53|6.33|
|0|0|WELL_512_A|0.02536|1.36|2.59|4.14|64.37|14.00|4.88|10.26|
|0|0|WELL_1024_A|0.03462|1.36|2.37|3.39|45.20|8.31|5.66|13.54|
|0|0|JDK|0.03498|0.79|1.93|2.34|4.54|2.82|0|3.68|
|0|0|XOR_SHIFT_1024_S|0.03962|1.28|2.10|2.60|75.25|12.26|3.55|7.24|
|0|0|MWC_256|0.46924|0.97|1.28|1.18|4.50|2.09|2.37|6.70|
|0|0|WELL_19937_C|0.49446|1.10|0.99|1.20|4.58|2.29|4.51|14.01|
|0|0|WELL_19937_A|0.53464|0.91|1.17|1.11|4.89|2.13|4.65|12.97|
|0|0|WELL_44497_B|1.10144|1.16|1.03|0.97|2.93|2.41|4.50|16.66|
|0|0|WELL_44497_A|1.189|0.98|0.89|1.06|3.06|1.87|5.32|12.29|
|0|0|ISAAC|2.46102|0.97|0.91|0.86|1.44|1.03|1.19|2.29|
|0|0|MT_64|7.13676|1.05|0.95|1.00|0.79|0.54|0.46|0.98|
|0|0|MT|11.00264|1.00|1.00|1.06|0.80|0.68|0.62|1.38|
|0|0|TWO_CMRES|133.73698|1.01|0.95|1.12|1.10|1.09|0|1.08|
|1|0.00398456|SPLIT_MIX_64|0.01308456|1.62|4.22|6.30|9.68|6.64|0|8.73|
|1|0.00603231|KISS|0.02735231|1.29|2.41|3.40|52.74|8.63|3.75|5.16|
|1|0.00735691|WELL_512_A|0.03271691|1.28|2.23|3.43|50.12|11.08|4.01|8.18|
|1|0.00744491|WELL_1024_A|0.04206491|1.30|2.13|2.97|37.38|7.02|4.84|11.32|
|1|0.01915509|JDK|0.05413509|0.87|1.60|1.86|3.29|2.18|0|2.73|
|1|0.00488982|XOR_SHIFT_1024_S|0.04450982|1.25|1.98|2.43|67.09|11.02|3.27|6.56|
|1|0.00413311|MWC_256|0.47337311|0.97|1.28|1.18|4.47|2.08|2.36|6.65|
|1|0.010572|WELL_19937_C|0.505032|1.09|0.99|1.20|4.51|2.26|4.44|13.74|
|1|0.01000549|WELL_19937_A|0.54464549|0.92|1.16|1.11|4.81|2.10|4.59|12.75|
|1|0.01212127|WELL_44497_B|1.11356127|1.16|1.03|0.97|2.91|2.40|4.46|16.49|
|1|0.01106587|WELL_44497_A|1.20006587|0.98|0.90|1.06|3.04|1.86|5.28|12.19|
|1|0.00597662|ISAAC|2.46699662|0.97|0.91|0.86|1.44|1.03|1.19|2.28|
|1|0.00619041|MT_64|7.14295041|1.05|0.95|1.00|0.79|0.54|0.46|0.98|
|1|0.00663587|MT|11.00927587|1.00|1.00|1.06|0.80|0.68|0.62|1.38|
|1|0.00458057|TWO_CMRES|133.74156057|1.01|0.95|1.12|1.10|1.09|0|1.08|
|10|0.0398456|SPLIT_MIX_64|0.0489456|1.16|1.86|2.42|3.32|2.51|0|3.07|
|10|0.0603231|KISS|0.0816431|1.10|1.47|1.80|18.34|3.56|1.92|2.39|
|10|0.0735691|WELL_512_A|0.0989291|1.09|1.41|1.81|17.24|4.33|1.99|3.37|
|10|0.0744491|WELL_1024_A|0.1090691|1.11|1.44|1.76|15.03|3.32|2.48|4.98|
|10|0.1915509|JDK|0.2265309|0.97|1.14|1.21|1.55|1.28|0|1.41|
|10|0.0488982|XOR_SHIFT_1024_S|0.0885182|1.13|1.49|1.72|34.23|6.04|2.14|3.79|
|10|0.0413311|MWC_256|0.5105711|0.98|1.26|1.17|4.22|2.00|2.26|6.24|
|10|0.10572|WELL_19937_C|0.60018|1.08|0.99|1.17|3.95|2.06|3.89|11.72|
|10|0.1000549|WELL_19937_A|0.6346949|0.93|1.14|1.10|4.27|1.95|4.08|11.08|
|10|0.1212127|WELL_44497_B|1.2226527|1.15|1.02|0.97|2.74|2.27|4.15|15.11|
|10|0.1106587|WELL_44497_A|1.2996587|0.98|0.90|1.05|2.89|1.79|4.95|11.33|
|10|0.0597662|ISAAC|2.5207862|0.97|0.91|0.86|1.43|1.03|1.19|2.26|
|10|0.0619041|MT_64|7.1986641|1.05|0.95|1.00|0.79|0.54|0.46|0.98|
|10|0.0663587|MT|11.0689987|1.00|1.00|1.06|0.80|0.68|0.62|1.38|
|10|0.0458057|TWO_CMRES|133.7827857|1.01|0.95|1.12|1.10|1.09|0|1.08|
|100|0.398456|SPLIT_MIX_64|0.407556|1.02|1.10|1.17|1.28|1.18|0|1.25|
|100|0.603231|KISS|0.624551|1.01|1.06|1.10|3.27|1.33|1.12|1.18|
|100|0.735691|WELL_512_A|0.761051|1.01|1.05|1.10|3.11|1.43|1.13|1.31|
|100|0.744491|WELL_1024_A|0.779111|1.02|1.06|1.11|2.96|1.33|1.21|1.56|
|100|1.915509|JDK|1.950489|1.00|1.02|1.02|1.06|1.03|0|1.05|
|100|0.488982|XOR_SHIFT_1024_S|0.528602|1.02|1.08|1.12|6.57|1.84|1.19|1.47|
|100|0.413311|MWC_256|0.882551|0.99|1.15|1.10|2.86|1.58|1.73|4.03|
|100|1.0572|WELL_19937_C|1.55166|1.03|1.00|1.06|2.14|1.41|2.12|5.15|
|100|1.000549|WELL_19937_A|1.535189|0.97|1.06|1.04|2.35|1.39|2.27|5.17|
|100|1.212127|WELL_44497_B|2.313567|1.08|1.01|0.99|1.92|1.67|2.66|8.45|
|100|1.106587|WELL_44497_A|2.295587|0.99|0.95|1.03|2.07|1.45|3.24|6.85|
|100|0.597662|ISAAC|3.058682|0.98|0.93|0.89|1.36|1.02|1.16|2.04|
|100|0.619041|MT_64|7.755801|1.04|0.96|1.00|0.80|0.58|0.50|0.98|
|100|0.663587|MT|11.666227|1.00|1.00|1.06|0.81|0.70|0.64|1.36|
|100|0.458057|TWO_CMRES|134.195037|1.01|0.95|1.12|1.10|1.09|0|1.08|
|10000|39.8456|SPLIT_MIX_64|39.8547|1.00|1.00|1.00|1.00|1.00|0|1.00|
|10000|60.3231|KISS|60.34442|1.00|1.00|1.00|1.02|1.00|1.00|1.00|
|10000|73.5691|WELL_512_A|73.59446|1.00|1.00|1.00|1.02|1.00|1.00|1.00|
|10000|74.4491|WELL_1024_A|74.48372|1.00|1.00|1.00|1.02|1.00|1.00|1.01|
|10000|191.5509|JDK|191.58588|1.00|1.00|1.00|1.00|1.00|0|1.00|
|10000|48.8982|XOR_SHIFT_1024_S|48.93782|1.00|1.00|1.00|1.06|1.01|1.00|1.01|
|10000|41.3311|MWC_256|41.80034|1.00|1.00|1.00|1.04|1.01|1.02|1.06|
|10000|105.72|WELL_19937_C|106.21446|1.00|1.00|1.00|1.02|1.01|1.02|1.06|
|10000|100.0549|WELL_19937_A|100.58954|1.00|1.00|1.00|1.02|1.01|1.02|1.06|
|10000|121.2127|WELL_44497_B|122.31414|1.00|1.00|1.00|1.02|1.01|1.03|1.14|
|10000|110.6587|WELL_44497_A|111.8477|1.00|1.00|1.00|1.02|1.01|1.05|1.12|
|10000|59.7662|ISAAC|62.22722|1.00|1.00|0.99|1.02|1.00|1.01|1.05|
|10000|61.9041|MT_64|69.04086|1.00|1.00|1.00|0.98|0.95|0.94|1.00|
|10000|66.3587|MT|77.36134|1.00|1.00|1.01|0.97|0.95|0.95|1.05|
|10000|45.8057|TWO_CMRES|179.54268|1.00|0.96|1.09|1.08|1.07|0|1.06|
|100000|398.456|SPLIT_MIX_64|398.4651|1.00|1.00|1.00|1.00|1.00|0|1.00|
|100000|603.231|KISS|603.25232|1.00|1.00|1.00|1.00|1.00|1.00|1.00|
|100000|735.691|WELL_512_A|735.71636|1.00|1.00|1.00|1.00|1.00|1.00|1.00|
|100000|744.491|WELL_1024_A|744.52562|1.00|1.00|1.00|1.00|1.00|1.00|1.00|
|100000|1915.509|JDK|1915.54398|1.00|1.00|1.00|1.00|1.00|0|1.00|
|100000|488.982|XOR_SHIFT_1024_S|489.02162|1.00|1.00|1.00|1.01|1.00|1.00|1.00|
|100000|413.311|MWC_256|413.78024|1.00|1.00|1.00|1.00|1.00|1.00|1.01|
|100000|1057.2|WELL_19937_C|1057.69446|1.00|1.00|1.00|1.00|1.00|1.00|1.01|
|100000|1000.549|WELL_19937_A|1001.08364|1.00|1.00|1.00|1.00|1.00|1.00|1.01|
|100000|1212.127|WELL_44497_B|1213.22844|1.00|1.00|1.00|1.00|1.00|1.00|1.01|
|100000|1106.587|WELL_44497_A|1107.776|1.00|1.00|1.00|1.00|1.00|1.00|1.01|
|100000|597.662|ISAAC|600.12302|1.00|1.00|1.00|1.00|1.00|1.00|1.01|
|100000|619.041|MT_64|626.17776|1.00|1.00|1.00|1.00|0.99|0.99|1.00|
|100000|663.587|MT|674.58964|1.00|1.00|1.00|1.00|0.99|0.99|1.01|
|100000|458.057|TWO_CMRES|591.79398|1.00|0.99|1.03|1.02|1.02|0|1.02|

The effect of construction and generation can be more easily observed using a table of the
relative contribution of the generation time to the total time. Here the total time is made
using construction with the *new* keyword or via the default *{{RandomSource.create}}* method
with a *native* seed or *null* seed. Note: Times for the *create* method are relative to
*new* (as above) but the generating % is computed using the total time.
||Values||Generation||RandomSource||new + Generation||Generating (%)||createNativeSeed (relative)||Generating
(%)||createNullSeed (relative)||Generating (%)||
|0|0|SPLIT_MIX_64|0.0091|0.00|8.62|0.00|13.48|0.00|
|0|0|KISS|0.02132|0.00|4.07|0.00|67.38|0.00|
|0|0|WELL_512_A|0.02536|0.00|4.14|0.00|64.37|0.00|
|0|0|WELL_1024_A|0.03462|0.00|3.39|0.00|45.2|0.00|
|0|0|JDK|0.03498|0.00|2.34|0.00|4.54|0.00|
|0|0|XOR_SHIFT_1024_S|0.03962|0.00|2.6|0.00|75.25|0.00|
|0|0|MWC_256|0.46924|0.00|1.18|0.00|4.5|0.00|
|0|0|WELL_19937_C|0.49446|0.00|1.2|0.00|4.58|0.00|
|0|0|WELL_19937_A|0.53464|0.00|1.11|0.00|4.89|0.00|
|0|0|WELL_44497_B|1.10144|0.00|0.97|0.00|2.93|0.00|
|0|0|WELL_44497_A|1.189|0.00|1.06|0.00|3.06|0.00|
|0|0|ISAAC|2.46102|0.00|0.86|0.00|1.44|0.00|
|0|0|MT_64|7.13676|0.00|1|0.00|0.79|0.00|
|0|0|MT|11.00264|0.00|1.06|0.00|0.8|0.00|
|0|0|TWO_CMRES|133.73698|0.00|1.12|0.00|1.1|0.00|
|1|0.00398456|SPLIT_MIX_64|0.01308456|30.45|6.3|4.83|9.68|3.15|
|1|0.00603231|KISS|0.02735231|22.05|3.4|6.49|52.74|0.42|
|1|0.00735691|WELL_512_A|0.03271691|22.49|3.43|6.56|50.12|0.45|
|1|0.00744491|WELL_1024_A|0.04206491|17.70|2.97|5.96|37.38|0.47|
|1|0.01915509|JDK|0.05413509|35.38|1.86|19.02|3.29|10.75|
|1|0.00488982|XOR_SHIFT_1024_S|0.04450982|10.99|2.43|4.52|67.09|0.16|
|1|0.00413311|MWC_256|0.47337311|0.87|1.18|0.74|4.47|0.20|
|1|0.010572|WELL_19937_C|0.505032|2.09|1.2|1.74|4.51|0.46|
|1|0.01000549|WELL_19937_A|0.54464549|1.84|1.11|1.66|4.81|0.38|
|1|0.01212127|WELL_44497_B|1.11356127|1.09|0.97|1.12|2.91|0.37|
|1|0.01106587|WELL_44497_A|1.20006587|0.92|1.06|0.87|3.04|0.30|
|1|0.00597662|ISAAC|2.46699662|0.24|0.86|0.28|1.44|0.17|
|1|0.00619041|MT_64|7.14295041|0.09|1|0.09|0.79|0.11|
|1|0.00663587|MT|11.00927587|0.06|1.06|0.06|0.8|0.08|
|1|0.00458057|TWO_CMRES|133.74156057|0.00|1.12|0.00|1.1|0.00|
|10|0.0398456|SPLIT_MIX_64|0.0489456|81.41|2.42|33.64|3.32|24.52|
|10|0.0603231|KISS|0.0816431|73.89|1.8|41.05|18.34|4.03|
|10|0.0735691|WELL_512_A|0.0989291|74.37|1.81|41.09|17.24|4.31|
|10|0.0744491|WELL_1024_A|0.1090691|68.26|1.76|38.78|15.03|4.54|
|10|0.1915509|JDK|0.2265309|84.56|1.21|69.88|1.55|54.55|
|10|0.0488982|XOR_SHIFT_1024_S|0.0885182|55.24|1.72|32.12|34.23|1.61|
|10|0.0413311|MWC_256|0.5105711|8.10|1.17|6.92|4.22|1.92|
|10|0.10572|WELL_19937_C|0.60018|17.61|1.17|15.06|3.95|4.46|
|10|0.1000549|WELL_19937_A|0.6346949|15.76|1.1|14.33|4.27|3.69|
|10|0.1212127|WELL_44497_B|1.2226527|9.91|0.97|10.22|2.74|3.62|
|10|0.1106587|WELL_44497_A|1.2996587|8.51|1.05|8.11|2.89|2.95|
|10|0.0597662|ISAAC|2.5207862|2.37|0.86|2.76|1.43|1.66|
|10|0.0619041|MT_64|7.1986641|0.86|1|0.86|0.79|1.09|
|10|0.0663587|MT|11.0689987|0.60|1.06|0.57|0.8|0.75|
|10|0.0458057|TWO_CMRES|133.7827857|0.03|1.12|0.03|1.1|0.03|
|100|0.398456|SPLIT_MIX_64|0.407556|97.77|1.17|83.56|1.28|76.38|
|100|0.603231|KISS|0.624551|96.59|1.1|87.81|3.27|29.54|
|100|0.735691|WELL_512_A|0.761051|96.67|1.1|87.88|3.11|31.08|
|100|0.744491|WELL_1024_A|0.779111|95.56|1.11|86.09|2.96|32.28|
|100|1.915509|JDK|1.950489|98.21|1.02|96.28|1.06|92.65|
|100|0.488982|XOR_SHIFT_1024_S|0.528602|92.50|1.12|82.59|6.57|14.08|
|100|0.413311|MWC_256|0.882551|46.83|1.1|42.57|2.86|16.37|
|100|1.0572|WELL_19937_C|1.55166|68.13|1.06|64.28|2.14|31.84|
|100|1.000549|WELL_19937_A|1.535189|65.17|1.04|62.67|2.35|27.73|
|100|1.212127|WELL_44497_B|2.313567|52.39|0.99|52.92|1.92|27.29|
|100|1.106587|WELL_44497_A|2.295587|48.20|1.03|46.80|2.07|23.29|
|100|0.597662|ISAAC|3.058682|19.54|0.89|21.95|1.36|14.37|
|100|0.619041|MT_64|7.755801|7.98|1|7.98|0.8|9.98|
|100|0.663587|MT|11.666227|5.69|1.06|5.37|0.81|7.02|
|100|0.458057|TWO_CMRES|134.195037|0.34|1.12|0.30|1.1|0.31|
|10000|39.8456|SPLIT_MIX_64|39.8547|99.98|1|99.98|1|99.98|
|10000|60.3231|KISS|60.34442|99.96|1|99.96|1.02|98.00|
|10000|73.5691|WELL_512_A|73.59446|99.97|1|99.97|1.02|98.01|
|10000|74.4491|WELL_1024_A|74.48372|99.95|1|99.95|1.02|97.99|
|10000|191.5509|JDK|191.58588|99.98|1|99.98|1|99.98|
|10000|48.8982|XOR_SHIFT_1024_S|48.93782|99.92|1|99.92|1.06|94.26|
|10000|41.3311|MWC_256|41.80034|98.88|1|98.88|1.04|95.07|
|10000|105.72|WELL_19937_C|106.21446|99.53|1|99.53|1.02|97.58|
|10000|100.0549|WELL_19937_A|100.58954|99.47|1|99.47|1.02|97.52|
|10000|121.2127|WELL_44497_B|122.31414|99.10|1|99.10|1.02|97.16|
|10000|110.6587|WELL_44497_A|111.8477|98.94|1|98.94|1.02|97.00|
|10000|59.7662|ISAAC|62.22722|96.05|0.99|97.02|1.02|94.16|
|10000|61.9041|MT_64|69.04086|89.66|1|89.66|0.98|91.49|
|10000|66.3587|MT|77.36134|85.78|1.01|84.93|0.97|88.43|
|10000|45.8057|TWO_CMRES|179.54268|25.51|1.09|23.41|1.08|23.62|
|100000|398.456|SPLIT_MIX_64|398.4651|100.00|1|100.00|1|100.00|
|100000|603.231|KISS|603.25232|100.00|1|100.00|1|100.00|
|100000|735.691|WELL_512_A|735.71636|100.00|1|100.00|1|100.00|
|100000|744.491|WELL_1024_A|744.52562|100.00|1|100.00|1|100.00|
|100000|1915.509|JDK|1915.54398|100.00|1|100.00|1|100.00|
|100000|488.982|XOR_SHIFT_1024_S|489.02162|99.99|1|99.99|1.01|99.00|
|100000|413.311|MWC_256|413.78024|99.89|1|99.89|1|99.89|
|100000|1057.2|WELL_19937_C|1057.69446|99.95|1|99.95|1|99.95|
|100000|1000.549|WELL_19937_A|1001.08364|99.95|1|99.95|1|99.95|
|100000|1212.127|WELL_44497_B|1213.22844|99.91|1|99.91|1|99.91|
|100000|1106.587|WELL_44497_A|1107.776|99.89|1|99.89|1|99.89|
|100000|597.662|ISAAC|600.12302|99.59|1|99.59|1|99.59|
|100000|619.041|MT_64|626.17776|98.86|1|98.86|1|98.86|
|100000|663.587|MT|674.58964|98.37|1|98.37|1|98.37|
|100000|458.057|TWO_CMRES|591.79398|77.40|1.03|75.15|1.02|75.88|

 

*Creating 1 value.*

Construction time dominates. For the fastest to construct (SplitMix) the time generating
the number is only 30% of the total time when using the *new* keyword, or 5% when constructing
with the native seed via the factory method.

The use case here is not to generate a single random number (instead use e.g. {{NumberFactory.nextLong()}})
but to have a {{UniformRandomProvider}} to pass to a method/object that requires one. Perhaps
an example should be added to the examples module that demonstrates how to do this using a
{{ThreadLocal<UniformRandomProvider>}} to provide thread safety for a single/low usage
RNG.

 

*Creating 10 values.*

The construction time is a noticable component of the total time. For the fastest to construct (SplitMix)
the time generating the number is now 80% of the total time when using the *new* keyword,
but mush lower at 33% when constructing with the native seed via the factory method. In this
use case it will be valid to optimise construction speed.

 

*Creating 100 values.*

For the fast to construct generators the generation time is the major component of the run-time.
All of these generators have either a primitive seed or a seed array with a size less than
32. The difference of using the {{RandomSource.create}} is noticeable and optimisation of
the construction will still be beneficial.

Compared to the above generating 100 values using a SplitMix64 takes 80% of the total time
when using the *new* keyword, or 84% when constructing with the native seed via the factory
method.

The slow to construct generators (MT, MT_64, TWO_CMRES) have internal seeding routines that
dominate total time.

The other generators MWC_256, Well family and ISAAC take 47%, 68-48% and 20% of the time generating
numbers. These have intermediate length seeds which are copied in the constructor adding overhead.

Note that there is a big difference between creating a generator with a native seed and with
no seed when not using a primitive seed as the cost to generate the seed is significant.

 

*Creating 10000 values.*

Only the slow to construct generators (MT, MT_64, TWO_CMRES) with internal seeding routines
part of the algorithm are still dominated by construction cost.

The cost of seed generation is now not very noticeable.

*Creating 1000000 values.*

By this point creating without a seed has no penalty verses with a seed.

Most generators spend nearly 100% of the time generating numbers.

The slow to construct MT and MT 64 spend <2% of the time constructing. Only the TWO_CMRES
still spends 24% of the time constructing the generator.

*It should be noted that generation is still fast. TWO_CMRES can be constructed from a random
seed and create 1,000,000 numbers in 592 microseconds, less than a millisecond.*

To preempt the question: Why bother if this is so fast anyway?
 A. Because a few small changes will significantly improve speed in the use case of constructing
short lifetime generators for use in for example array shuffling. And "[because it's there|https://blog.theclymb.com/out-there/because-its-there-the-quotable-george-mallory/]".
  

Improvements ideas:
 * Generators with an array seed construct a new array (zero filled) and call the method to
fill the array using the input seed. In the case where the input seed is the correct size
then seed filling can be skipped and the seed copied using a call to {{Arrays.copyOf}} which
should be faster as it is JVM intrinsic and avoids zero fill.
 * The MWC_256 is interesting as it has a seed size of 257. However the construction creates
a temp array which is filled and then later copied to the state which has a size of 256. If
the input seed is the correct size then this allocation for seed filling is redundant. This
provides an optimisation.

> Create a RandomSource.create benchmark
> --------------------------------------
>
>                 Key: RNG-72
>                 URL: https://issues.apache.org/jira/browse/RNG-72
>             Project: Commons RNG
>          Issue Type: New Feature
>          Components: simple
>    Affects Versions: 1.3
>            Reporter: Alex D Herbert
>            Assignee: Alex D Herbert
>            Priority: Minor
>              Labels: performance-benchmark
>             Fix For: 1.3
>
>          Time Spent: 20m
>  Remaining Estimate: 0h
>
> The recommended method to construct a {{UniformRandomProvider}} is to use, e.g.:
> {code:java}
> import org.apache.commons.rng.UniformRandomProvider;
> import org.apache.commons.rng.simple.RandomSource;
> UniformRandomProvider rng = RandomSource.create(RandomSource.MWC_256);
> {code}
> The factory method knows the type of seed required for the constructor and generates
one as appropriate.
> This factory method could be made more efficient, in particular:
>  * Reducing synchronisation around the single source of random seed data
>  * Adding knowledge of the required seed size for arrays
>  * Changing internal data structures, e.g. {{Map<Class<?>, SeedConverter<?,?>>}}
can be changed to {{Map<SeedType, SeedConverter<?,?>>}} using an {{EnumMap}} if
a new enum {{SeedType}} was created for all the supported seeds (currently 4 types).
>  * Add a new interface to replace {{SeedConverter<?,?>.convert()}} with a {{.convert(int
outputArraySize)}} method to allow conversions to generate appropriately sized arrays. The
parameter can be ignored for non-array conversions but could optimise array conversions.
> This ticket is to add a JMH benchmark to compare the speed of construction of all the
providers using:
>  * Their native constructor
>  * {{RandomSource}} using the native seed of the correct size (calls a constructor using
reflection)
>  * {{RandomSource}} using a non native seed (requires seed conversion)
>  * {{RandomSource}} using no seed (requires seed generation)
> The report will be posted here. It could be added to the user guide for reference.
> This work is motivated by the new {{XorShiRo}} generators in version 1.3 that have a
native array seed size of 2, 4, or 8. The current {{RandomSource}} create method will generate
a fixed seed of length 128 for seeding.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message