BPF Benchmark

This repository now follows a three-layer layout for the active benchmarking pipeline:

• micro/: isolated micro-benchmarks driven by bpf_prog_test_run_opts
• corpus/: real-world program collection, measurement, and analysis
• e2e/: end-to-end deployment benchmarks and case-specific assets

Legacy material that is no longer part of the active pipeline lives under legacy/.

Repository Layout

bpf-benchmark/
├── micro/                 # Isolated micro-benchmark suites + shared orchestrator
├── corpus/                # Real-world corpus configs, fetch/build, measurement, analysis
├── e2e/                   # End-to-end workloads, case assets, and result placeholders
├── legacy/                # Inactive historical benchmark code and third-party snapshots
├── docs/                  # Research plans, reports, and temporary experiment notes
├── scanner/               # Offline directive scanner
└── vendor/                # Vendored kernel / tooling dependencies

Quick Start

Initialize submodules, activate the shared virtualenv used in this workspace, and build the micro layer:

git submodule update --init --recursive
source /home/yunwei37/workspace/.venv/bin/activate
make -C micro

List the current micro suite and run the minimal smoke benchmark:

python3 micro/run_micro.py --list
python3 micro/run_micro.py --bench simple --runtime llvmbpf --iterations 1 --warmups 0 --repeat 10
python3 micro/run_micro.py --bench simple --runtime kernel --iterations 1 --warmups 0 --repeat 10

The kernel runtime command requires passwordless sudo -n.

Build the standalone scanner CLI required by corpus/ recompile paths and e2e/:

cmake -S scanner -B scanner/build -DCMAKE_BUILD_TYPE=Release
cmake --build scanner/build --target bpf-jit-scanner -j

Check the higher-level entrypoints:

python3 micro/driver.py suite -- --list
python3 corpus/build_expanded_corpus.py --help
python3 e2e/run.py --help

Prepare the corpus-derived objects used by the end-to-end cases:

python3 corpus/build_expanded_corpus.py --repo tracee --repo tetragon --repo scx

Then run layer-specific workflows:

python3 micro/driver.py suite -- --runtime llvmbpf --runtime kernel
python3 corpus/directive_census.py --help
python3 corpus/run_macro_corpus.py --list
python3 e2e/run.py tracee --smoke
python3 e2e/run.py tetragon --smoke
python3 e2e/run.py bpftrace --smoke
python3 e2e/run.py scx --smoke

Layer Notes

micro/ owns the isolated benchmark manifests, runner build, and the shared micro/orchestrator/ Python utilities that corpus/ and e2e/ import.

corpus/ owns the 23-project real-world corpus, fetch/build tooling, declarative corpus config in corpus/config/, and the measurement / analysis scripts for corpus-wide experiments.

e2e/ owns full deployment-style evaluation via the unified e2e/run.py entrypoint plus per-case assets under e2e/cases/.

legacy/ contains the old multi-runtime userspace benchmark and external reference repositories. They are preserved for history but are not part of the active refactored workflow.