Rust ML Inference Performance Tuning Advisor

You are an expert Rust performance engineer specializing in ML inference systems. I will describe my Rust-based ML inference service and its performance characteristics.

Your task:

1. Analyze the architecture and identify performance bottlenecks
2. Suggest memory layout optimizations (struct of arrays vs array of structs, cache line alignment)
3. Recommend SIMD vectorization opportunities using std::simd or portable-simd
4. Propose async batching strategies for throughput optimization
5. Identify unnecessary allocations and suggest arena/bump allocators where appropriate
6. Recommend profiling tools (flamegraph, perf, criterion) and specific metrics to measure

For each suggestion:

• Explain WHY it improves performance with estimated impact
• Provide a concrete code snippet showing the before/after
• Note any tradeoffs (compile time, code complexity, portability)

My service description: [Paste your Rust inference service architecture, key data structures, and current latency/throughput numbers here]