A collection of JMH benchmarks which may be useful for measuring Armeria performance.

• -Pjmh.include=<pattern>
  • The benchmarks to run, in a comma-separated regular expression. All benchmarks if unspecified.
    • grpc.downstream.DownstreamSimpleBenchmark
    • grpc.downstream.DownstreamSimpleBenchmark.empty$
• -Pjmh.params=<spec>
  • The benchmark parameters. Uses the parameters specified in the benchmark code if unspecified.
    • clientType=NORMAL
    • num=10,100;flowControl=false
• -Pjmh.fork=<integer>
  • The number of forks. 1 if unspecified.
• -Pjmh.iterations=<integer>
  • The number of iterations. JMH default if unspecified.
• -Pjmh.warmupIterations
  • The number of iterations. Uses the value of jmh.iterations if unspecified.
• -Pjmh.profilers=<spec>
  • The profiler settings. Profiler disabled if unspecified.
    • jmh.extras.Async:asyncProfilerDir=...;flameGraphDir=...
• -Pjmh.threads=<integer>
  • The number of threads. JMH default if unspecified.
• -Pjmh.verbose
  • Increases the verbosity of JMH to EXTRA.
• -Pjmh.jvmargs=<jvm options>
  • Additional JVM options.
    • -Xmx8192m -Xms8192m
• -Pjmh.forcegc=<true|false>
  • Whether to force JVM garbage collection. false if unspecified.

Allow running perf as a normal user:

# echo 1 > /proc/sys/kernel/perf_event_paranoid
# echo 0 > /proc/sys/kernel/kptr_restrict

Install async-profiler and FlameGraph:

$ cd "$HOME"
$ git clone https://github.com/jvm-profiling-tools/async-profiler.git
$ git clone https://github.com/brendangregg/FlameGraph.git
$ cd async-profiler
$ make

When running a benchmark, specify -Pjmh.profilers option:

$ ./gradlew :benchmarks:jmh \
  "-Pjmh.profilers=jmh.extras.Async:asyncProfilerDir=$HOME/async-profiler;flameGraphDir=$HOME/FlameGraph"

• Do not forget to wrap -Pjmh.params and -Pjmh.profilers option with double quotes, because otherwise your shell will interpret ; as the end of the command.
• See sbt-jmh documentation for more profiler options.