false-sharing

The repository to demonstrate the impact of false sharing on performance.

Blog post about false-sharing

Benchmark results

Two classes were compared. One class was affected by false sharing, the other one uses @Contended annotation in order to solve the issue.

Class affected by false sharing:

Benchmark                                             Mode  Cnt   Score   Error  Units
FalseSharingBenchmark.falseSharing                    avgt   20  48,897 ± 0,763  ns/op
FalseSharingBenchmark.falseSharing:incrementCounterA  avgt   20  48,905 ± 0,898  ns/op
FalseSharingBenchmark.falseSharing:incrementCounterB  avgt   20  48,889 ± 0,859  ns/op

After @Contended was added to the class:

Benchmark                                             Mode  Cnt   Score   Error  Units
FalseSharingBenchmark.contended                       avgt   20   6,765 ± 0,063  ns/op
FalseSharingBenchmark.contended:contendedIncrementA   avgt   20   6,765 ± 0,063  ns/op
FalseSharingBenchmark.contended:contendedIncrementB   avgt   20   6,765 ± 0,063  ns/op

Build

mvn clean package
docker build -t false-sharing .

Run

You can run benchmarks by yourself. Docker container shows benchmark results and perf output. Perf is a Linux profiler and it's able to show the number of CPU cache load events.

Benchmark class affected by false-sharing:

 docker run --cap-add SYS_ADMIN \
  false-sharing \
  java -jar -XX:-RestrictContended /opt/app/app.jar \
  FalseSharingBenchmark

Benchmark class with @Contended annotation:

 docker run --cap-add SYS_ADMIN \
  false-sharing \
  java -jar -XX:-RestrictContended /opt/app/app.jar \
  ContendedBenchmark

Run all benchmarks:

 docker run --cap-add SYS_ADMIN \
  false-sharing \
  java -jar -XX:-RestrictContended /opt/app/app.jar