The author details their journey in optimizing a Rust code using the rayon framework. Initially disappointed by only a 2x speedup on an 8 CPU thread computer, they explore profiling tools like strace and perf to identify bottlenecks. Surprisingly, the author discovers that copying data instead of zero-copy structures improves performance. The use of flame graphs in the Firefox Profiler provides insights into function calls and system calls, revealing cache misses as crucial for performance. The importance of optimizing CPU cache usage is highlighted, as small cache sizes can significantly impact performance. Stay tuned for more optimization strategies in upcoming posts.
https://gendignoux.com/blog/2024/11/18/rust-rayon-optimized.html