Skip to content

README.md

Latest commit

 

History

History
71 lines (48 loc) · 3.14 KB

README.md

File metadata and controls

71 lines (48 loc) · 3.14 KB

Firewood Benchmarks

Firewood has two categories of benchmarks:

  • Macro benchmarks — C-Chain re-execution against real mainnet state; measures end-to-end system throughput
  • Micro benchmarks — synthetic workloads and criterion suites that exercise the Rust API directly, no AvalancheGo required

What we have

Benchmark Use when Environment
C-Chain re-execution — GitHub Actions Tracking performance over time, A/B between versions Managed CI — scheduled, results published to GitHub Pages
C-Chain re-execution — fwdctl launch Investigating regression, in-depth profiling, tuning EC2 instance you provision — full root access
Rust criterion Iterating on a specific operation locally; enforces pure-Rust API in CI Local or CI (benchmarks.yaml)
Synthetic workloads Testing Firewood API patterns without AvalancheGo Local

Details

C-Chain re-execution — GitHub Actions

When: tracking performance over time or comparing two versions — runs automatically on a daily schedule and on every trigger, no manual setup needed.

Part of CI — repeatable, versioned, and auditable. Runs on a fixed schedule and on demand via track-performance.yml. Each run is isolated on a dedicated self-hosted runner, keeping variance low enough that a meaningful difference reflects code, not infrastructure. Results accumulate on GitHub Pages.

→ Full guide

C-Chain re-execution — fwdctl launch

When: a GitHub Actions run surfaced a signal worth investigating, or when running reexecution tests expected to take longer than 24 hours — SSH into the instance, install any tooling, change code and rebuild freely.

Same workload as GitHub Actions, provisioned on demand on EC2. Full root access — install perf, flamegraphs, or any tooling, change code and rebuild freely. No CI queue, no constraints.

→ Full guide

Rust criterion

When: changing a core area of Firewood — these benchmarks cover small critical sections and are the fastest way to detect performance problems before introducing the FFI and Go layers.

Criterion benchmarks live in firewood/benches/ and storage/benches/. They run in seconds with no external dependencies and also run in CI on every push to main.

Changed area Benchmark file
Deferred persistence firewood/benches/defer_persist.rs
Hashing firewood/benches/hashops.rs
Node serialization / deserialization storage/benches/serializer.rs 
cargo bench --features ethhash,logger

Benchmarks can also produce flamegraphs. See the header comment in firewood/benches/hashops.rs for instructions; --profile-time=5 is a good starting point.

Synthetic workloads

When: changing any core area of Firewood — these tests can help identify performance problems before introducing the FFI and Go layers.

A standalone Rust binary exercising the Firewood API directly with synthetic trie patterns (tenkrandom, zipf, single). No AvalancheGo, no Go, no cloud.

→ Workload specs