Skip to content

BENCHMARKING.md

Latest commit

 

History

History
99 lines (62 loc) · 2.81 KB

BENCHMARKING.md

File metadata and controls

99 lines (62 loc) · 2.81 KB

Running JMH Benchmarks in Besu

Besu includes JMH microbenchmarks for performance testing. This guide explains how to build and run them using Gradle.

🛠️ Prerequisites

  • Java 25+ (ensure JAVA_HOME is set)
  • Gradle (you can use the wrapper: ./gradlew)
  • Optional: Async Profiler for low-overhead profiling

🏃 Running Benchmarks

JMH tasks are defined in various modules such as ethereum:core, ethereum:eth, and ethereum:rlp.

▶️ Run All Benchmarks

./gradlew :ethereum:core:jmh

This runs all available benchmarks in the core module using default settings.

Gradle won't rerun a task by default with no changes, so to run subsequent times without changes add the gradle --rerun-tasks option.

🔁 Rerun All Benchmarks

./gradlew :ethereum:core:jmh --rerun-tasks

🎯 Filter Benchmarks by Name and Case

To run a specific benchmark class, use the -Pincludes and/or -Pexcludes project properties. To filter by case name, use -Pcases:

./gradlew :ethereum:core:jmh -Pincludes=Mod -Pexcludes=Mul,Add,SMod -Pcases=MOD_256_128,MOD_256_192 --rerun-tasks

This uses a regex pattern so other kinds of regexes can be used.

⚙️ Benchmark Configuration Options

For other configuration options run:

./gradlew help --task jmh

📦 Module-Specific Benchmarks

engine_getPayloadBodies parallelization (ethereum:api)

Benchmarks sequential vs parallel block body DB lookups for engine_getPayloadBodiesByHash{V1,V2} and engine_getPayloadBodiesByRange{V1,V2}. Uses simulated per-lookup latency to model warm-cache (100µs) and cold-cache (500µs) RocksDB reads.

./gradlew :ethereum:api:jmh -Pincludes=EngineGetPayloadBodiesParallel --rerun-tasks --no-daemon

🔥 Async Profiler Integration (Optional)

To profile benchmarks with Async Profiler:

./gradlew :ethereum:core:jmh \
  -Pincludes=SomeBenchmark \
  -PasyncProfiler=/path/to/libasyncProfiler.so \
  -PasyncProfilerOptions="output=flamegraph" \
  --rerun-tasks

This will generate two html profiling files flame-cpu-forward.html and flame-cpu-reverse.html after the benchmark run.

🧪 Sample Output

Benchmark                                Mode  Cnt  Score   Error  Units
SomeBenchmark.benchmark                  avgt   20  3.893 ± 0.021  ns/op
SomeBenchmark.benchmark                  avgt   20  8.951 ± 0.149  ns/op

📚 References