Phase 9 Performance Baseline

English counterpart of PERFORMANCE_BASELINE.md. | Türkçe

Environment

• Date: 2026-08-18
• Java: Temurin/OpenJDK 25.0.4
• Clojure: 1.12.5
• Operating system: Windows
• Commands: clojure -M:benchmark, clojure -M:soak
• Network: both tests run entirely in-process; they do not measure Binance latency.

Each microbenchmark performs 30 × 1,000 measured operations after 5,000 warmup operations. p50/p95 are calculated from nanoseconds-per-operation samples for each batch. Results are release regression references and vary with CPU, JVM warmup, and system load.

Microbenchmark

Scenario	p50 ns/op	p95 ns/op	Approx. throughput/sec
Bounded buffer offer/poll	995	1,751	1,004,823
HMAC WebSocket signing	125,442	176,845	7,971
Market JSON parse + normalize	11,205	16,012	89,243
User event normalize	8,864	10,312	112,812

Short Soak

A ten-second single-producer/single-consumer loop combined executionReport normalization with bounded queue offer/poll:

Measurement	Result
Events processed	1,383,948
Messages/sec	138,394
Final queue depth	0
Dropped events	0
Thread delta	0
Post-GC heap delta	2,024 bytes

Interpretation and Limits

• This local baseline shows capacity well above expected manual-trading bot load; it is not a production-throughput guarantee.
• HMAC measurement excludes signer construction and measures one payload signature with a reusable client signer.
• REST latency depends heavily on the Internet path and Testnet load. One acceptance run cannot define an SLA.
• RSS/CPU/GC profiling is outside this short dependency-free gate. Run long production soak and profiler work in the bot deployment environment.
• A p95 increase above 2× on the same machine, or any soak drop/thread growth, warrants investigation; no rigid automatic performance threshold is enforced.