Identify and eliminate unnecessary memory allocations in performance-critical code paths. This includes pre-allocating collections when the size is known, using static strings for common values, and avoiding intermediate data structures. Key strategies: - **Pre-allocate collections**: When you know the expected size, pre-allocate vectors and hashmaps to avoid repeated reallocations - **Use static strings**: For common values like HTTP methods, use `Cow::Borrowed` or `FastStaticString` instead of allocating new strings - **Eliminate intermediate collections**: Avoid creating temporary data structures when you can process data in a single pass - **Reuse attribute objects**: Cache and reuse expensive-to-create objects rather than recreating them Example from HTTP method handling: ```rust // Instead of always allocating: fn method_slow(method: &http::Method) -> String { method.to_string() // Always allocates } // Use static strings for common cases: fn method_fast(method: &http::Method) -> Cow<'static, str> { match *method { Method::GET => Cow::Borrowed("GET"), Method::POST => Cow::Borrowed("POST"), // ... other common methods _ => Cow::Owned(method.to_string()), } } ``` This optimization is particularly important in hot code paths where even small allocations can compound into significant performance overhead. Always measure the impact, as the complexity trade-off should be justified by measurable performance gains.