Local Setup

The AI Gateway has been tested under high load conditions to ensure optimal performance. This guide explains how to run benchmarks and interpret the results.

​

Running Benchmarks

There are two methods to benchmark the gateway:

​

1. Automated Benchmark Script

Use our automated benchmark script for comprehensive testing:

./scripts/benchmark.sh

The script automatically:

• Verifies dependencies
• Creates a test gateway
• Configures forwarding rules
• Executes multiple benchmark scenarios

​

Configuration

Customize the benchmark using environment variables:

export ADMIN_URL="http://localhost:8080/api/v1"  # Admin API endpoint
export PROXY_URL="http://localhost:8081"         # Proxy endpoint
export CONCURRENT_USERS=50                       # Number of concurrent users
export DURATION="30s"                           # Test duration

​

2. Manual Testing

For custom benchmarks, use hey:

1. Install hey:

go install github.com/rakyll/hey@latest

2. Run custom benchmarks:

# Set your gateway URL and API key
export GATEWAY_URL="http://localhost:8080"
export API_KEY="your-api-key"

# Run benchmark
hey -n 100000 -c 50 \
    -H "X-TG-API-Key: $API_KEY" \
    -H "Content-Type: application/json" \
    -m POST \
    -d '{"model": "gpt-3.5-turbo", "messages": [{"role": "user", "content": "Hi"}]}' \
    "$GATEWAY_URL/v1/chat/completions"

​

Performance Results

Recent benchmark results show impressive performance:

Summary:
  Total:        3.9210 secs
  Slowest:      0.1347 secs
  Fastest:      0.0004 secs
  Average:      0.0078 secs
  Requests/sec: 25503.4763

  Total data:   4900000 bytes
  Size/request: 49 bytes

Latency distribution:
  10% in 0.0047 secs
  25% in 0.0059 secs
  50% in 0.0073 secs
  75% in 0.0090 secs
  90% in 0.0109 secs
  95% in 0.0123 secs
  99% in 0.0158 secs

​

Key Metrics

​

Performance Characteristics

The gateway demonstrates:

• Low Latency: 90% of requests complete under 11ms
• High Throughput: Exceeds 25k requests per second
• Stability: Minimal performance variance
• Reliability: No failed requests during testing

​

Notes

• Results were obtained in controlled conditions
• Actual performance may vary based on:
  • Network conditions
  • Hardware specifications
  • Specific use cases
  • Load patterns

​

Best Practices

1. Environment Setup

   • Use production-grade hardware
   • Ensure sufficient network bandwidth
   • Configure appropriate system limits

2. Test Configuration

   • Start with lower concurrency
   • Gradually increase load
   • Monitor system resources

3. Results Analysis

   • Focus on P95 and P99 latencies
   • Monitor error rates
   • Check resource utilization