How do you minimize garbage collection pauses or runtime overhead in systems handling millions of requests per second?
