Performance is treated as a security feature in AetherCore. Slow systems enable Denial-of-Service vectors by exhausting connection pools and memory limits.
Below are the benchmark metrics continuously verified by our GitHub Actions pipeline (benchmark.sh).
Tested on Linux AMD64 GitHub Actions Runner (2 vCPU)
| Operation | LangChain (Python) | AetherCore (Go) | Measurement Bound |
|---|---|---|---|
| Cold Boot (CLI to Ready) | ~3,500ms | <10ms | Binary exec to main() |
| Subsystem Init + Log Config | N/A | <5ms | Slog + OS Signaling |
| IPC Ping/Pong (Rust Sidecar) | N/A | <1ms | Local Domain Socket |
AetherCore restricts garbage collector churn during heavy task loads by pooling memory for event evaluation.
| Benchmark Suite | Allocations per Op (allocs/op) |
Memory per Op (B/op) |
|---|---|---|
BenchmarkTaskDispatch |
0 allocs/op | 0 B/op |
BenchmarkJSONLogOutput |
0 allocs/op | 0 B/op |
BenchmarkCronMatcher |
0 allocs/op | 0 B/op |
Note: By hitting mathematically 0 allocs/op during the core loop dispatch, the inner Event Loop does not pressure the Go Garbage Collector, delivering deterministic latency percentiles.
The AetherCore application is distributed as a single statically-linked binary, heavily pruned of standard framework bloat.
- Target Size:
<10 MB - Stripped: Yes (
-s -w)
To replicate these benchmarks on your local machine:
make benchmark