This project measures timer latency (average time spent on timer call)
and granularity (average time in for which timer value stays the same)
of System.nanoTime() and System.currentTimeMillis() in a few modes.
This gives the estimates on the timer precision used for performance
assessment. I could do this experiment internally, but figured other
guys would find this data useful as well.
Build it from source:
$ mvn clean package
Run it:
$ java -jar target/timerbench.jar
Sample output:
---- 8< (cut here) -----------------------------------------
Java(TM) SE Runtime Environment, 1.8.0-b132
Java HotSpot(TM) 64-Bit Server VM, 25.0-b70
Linux, 3.13.0-custom+, amd64
Burning up to figure out the exact CPU count....... done!
Running with 1 threads and [-client]:
granularity_currentTime: 989264.177 +- 1769.248 ns
granularity_nanotime: 35.077 +- 0.131 ns
latency_currentTime: 42.370 +- 2.374 ns
latency_nanotime: 35.430 +- 0.107 ns
Running with 2 threads and [-client]:
granularity_currentTime: 993246.791 +- 1523.273 ns
granularity_nanotime: 49.965 +- 3.673 ns
latency_currentTime: 51.368 +- 3.529 ns
latency_nanotime: 42.261 +- 4.616 ns
Running with 4 threads and [-client]:
granularity_currentTime: 1005131.127 +- 6633.599 ns
granularity_nanotime: 56.744 +- 2.014 ns
latency_currentTime: 59.416 +- 2.466 ns
latency_nanotime: 54.162 +- 1.247 ns
Running with 1 threads and [-server]:
granularity_currentTime: 989051.057 +- 1759.073 ns
granularity_nanotime: 35.422 +- 0.488 ns
latency_currentTime: 40.839 +- 0.103 ns
latency_nanotime: 35.457 +- 0.340 ns
Running with 2 threads and [-server]:
granularity_currentTime: 993251.781 +- 899.136 ns
granularity_nanotime: 45.761 +- 4.810 ns
latency_currentTime: 48.669 +- 4.069 ns
latency_nanotime: 40.424 +- 4.293 ns
Running with 4 threads and [-server]:
granularity_currentTime: 1000651.038 +- 4507.238 ns
granularity_nanotime: 55.954 +- 1.864 ns
latency_currentTime: 59.752 +- 1.937 ns
latency_nanotime: 54.484 +- 4.416 ns
---- 8< (cut here) -----------------------------------------