Debug a slow concurrent network server under load

## Scenario You are on-call for a high-throughput network service (e.g., a TCP/HTTP server). Under load, users report: - Throughput suddenly drops - p99 latency increases significantly - CPU may be high or low depending on the incident The server uses a concurrency model based on an **event loop** and OS I/O multiplexing (e.g., `epoll`), and may use a thread pool for some work. ## Task Walk through how you would **debug and improve performance**. Your interviewer may ask about: - Connection handling and backlog - Event loop behavior - Blocking vs non-blocking I/O - I/O multiplexing (`select/poll/epoll`, edge vs level triggered) - Where `io_uring` might help vs where it would not - Typical bottlenecks: syscalls, context switching, lock contention, head-of-line blocking, buffer management ## Output expected Explain a structured approach: 1. What you would measure/collect first 2. How you would narrow down the bottleneck 3. Concrete fixes and trade-offs 4. How you would validate improvements and avoid regressions