This article recounts a two‑week Linux system tuning project that reduced a high‑traffic API’s P99 response time from over 500 ms to under 100 ms by systematically diagnosing bottlenecks, applying USE‑based analysis, and tuning kernel, network, file‑descriptor, memory, CPU, and I/O parameters.
This guide walks through real‑world Linux high‑concurrency scenarios, diagnosing TCP state bottlenecks, analyzing default kernel parameters, and applying practical sysctl tweaks, queue and buffer optimizations, keepalive settings, and monitoring scripts to dramatically improve connection handling and throughput.
This article details Ctrip's mobile app network service architecture, the rationale for using TCP over HTTP, and a series of channel governance and performance optimization techniques—including DNS bypass, socket connection pooling, weak‑network handling, data format improvements, retry mechanisms, Hybrid and overseas network enhancements—demonstrating how these measures raised service success rates above 99% and reduced latency.
