Skip to main content

​
Benchmark Suite

RotorTree includes a comprehensive benchmark suite with ~380 benchmarks covering different workload patterns and configurations. All benchmarks are driven by divan and crabtime for reliable performance measurements.

​
Test Environment

All benchmark results documented here were collected on:
  • CPU: Apple M4 Pro (14 cores)
  • Memory: 48 GB
  • OS: macOS
Your results may vary based on hardware, OS, and system load. Always benchmark your specific workload on your target environment.

​
Benchmark Categories

The benchmark suite is organized into four main categories:

In-Memory

Single-threaded and parallel insertion benchmarks for pure in-memory trees

Storage

WAL persistence, checkpoint, recovery, and mmap tiering benchmarks

Concurrent

Concurrent read/write workloads with lock-based coordination

​
Branching Factors

All benchmarks test four branching factors: N=2, 4, 8, 16. The branching factor affects:
  • Tree depth (higher N = shallower tree)
  • Hash computation per level (higher N = more hashes per node)
  • Storage layout (higher N = better disk block utilization)
  • Proof size (higher N = fewer siblings per level)

​
Measurement Approach

​
Throughput

Insertion benchmarks report throughput as leaves/second. Each benchmark:
  • Measures time to insert a specific number of leaves
  • Uses divan::counter::ItemsCount to calculate items/sec
  • Excludes setup time (tree creation, temp directory allocation)

​
Latency

Proof generation and verification benchmarks measure latency per operation. See the proof latency graph for mmap vs no-mmap comparison.

​
Variance

The pure in-memory benchmark (tree_bench_parallel) exhibits low variance and achieves peak throughput up to ~190M leaves/sec.The single-threaded variant (tree_bench) has the best variance characteristics — useful when predictability under load is a constraint, though at lower absolute throughput.

​
Interactive Results

Full benchmark results with detailed charts and comparison tools are available at: https://rymnc.github.io/rotortree/benchmarks The interactive dashboard includes:
  • Time-series performance trends
  • Branching factor comparisons
  • Workload scaling analysis
  • Statistical variance visualization

​
Benchmark Source Code

All benchmarks live in benches/ in the RotorTree repository:
  • tree_bench.rs - Single-threaded in-memory
  • tree_bench_parallel.rs - Parallel in-memory (rayon)
  • tree_bench_storage.rs - WAL persistence and recovery
  • tree_bench_concurrent.rs - Concurrent read/write workloads
  • tree_bench_all.rs - Combined suite
Run locally with: 
cargo bench -- --list  # list all benchmarks
cargo bench            # run all benchmarks

Build docs developers (and LLMs) love

Get started for freeTalk to us