✓ Criterion v0.5 · Apple Silicon
Benchmarks
Statistical benchmarks from Criterion with Wasmtime 18.0 release builds. Every number includes full host↔guest memory bridge overhead.
Parse Throughput
| Payload | Median | Throughput | Typical Use |
|---|
| 1 KB | 12.7 μs | 78.7 MB/s | Modbus RTU Frame |
| 10 KB | 12.9 μs | 775 MB/s | OPC UA ReadResponse |
| 100 KB | ~15 μs | ~6.5 GB/s | EDIFACT BAPLIE |
Cold Start
| Operation | Mean | Description |
|---|
| Module::new() | 0.35 ms | JIT compile from .wasm bytes |
| Module::deserialize() | 0.11 ms | Memory-map from .cwasm |
| Full Pool Create | 4.41 ms | Engine + compile + first instance |
| Pre-compiled Pool | 0.50 ms | Engine + deserialize + first instance |
Runtime Comparison
1KB payload parse latency
| Runtime | Latency | Safety | Binary | Isolation |
|---|
| VECSLATE | 12.7 μs | ✓ Compile | 77 KB | ✓ Sandbox |
| Node.js | ~250 μs | GC | ~45 MB | ✗ Process |
| Python | ~1,000 μs | GC | ~90 MB | ✗ Process |
Methodology
Statistical benchmarks via Criterion v0.5 with async Tokio runtime.
Parse path measures the full critical path: pool checkout → Wasm alloc → memcpy → parse → result read → dealloc.
Reproduce locally: cargo bench -p vecslate-core
← Driver Catalog