Textbooks tell us that read-write locks allow concurrent readers without blocking. On real multi-core hardware, adding cores collapsed throughput by 74%. Here is the physics of why, and how we achieved an 84x speedup with Optimistic Lock Coupling.
Hey, I’m Mohit Kumar.
Iām a backend engineer, and this is my corner of the internet where I write while learning.
Most posts are about systems, distributed systems, performance, databases, and code-level experiments. Sometimes I also write random things that I find interesting.
Recently published
Recent Writings
- Read-Locks Are Writes in Disguise: Achieving 84x Concurrency on Apple M2
- Making Binary Search Faster on Apple M2
Binary search is O(log N) and standard library code like std::lower_bound is heavily optimized. Can we make it nearly 3x faster on Apple Silicon? Here is what happens when you dismantle branch mispredictions and cache lines.
- Unknown Place
What is this place? Confused (visibly) š
Thanks for stopping by, stranger!