Improve concurrency beyond a single lock

You are given a simple multi-threaded component that protects all shared state with a single mutual-exclusion lock (mutex).

Scenario

A service has many worker threads calling get(key) and put(key, value) on a shared in-memory key–value store (e.g., a hash map). The current implementation uses one global mutex around the entire get/put logic to ensure correctness.

Tasks

1. Explain why the single global lock is correct but can hurt performance under contention.
2. Propose concrete ways to improve concurrency / throughput while keeping correctness. Discuss trade-offs.
3. Call out common pitfalls (deadlocks, race conditions, starvation) and how you would test/validate the improved design.