One Linux Kernel Parameter Was Throttling 11 of 14 Test Phones to 15 Mbps

A developer investigating inconsistent USB tethering speeds discovered that 11 out of 14 test phones were capped at exactly 15 Mbps, while three others reached 106 Mbps with no vendor documentation explaining the gap. The investigation began around 2:47 AM during an unrelated CI failure, prompting the engineer to build a custom diagnostic tool from scratch rather than rely on existing dashboards. The tool was written as a single Python file using only standard library modules — asyncio, collections, and dataclasses — avoiding heavy Node.js or npm dependency stacks that would have consumed hundreds of megabytes of RAM. A key performance fix involved eliminating repeated list conversions inside a sliding-window loop, reducing per-run memory allocation overhead from roughly 200ms to 2ms and eliminating GC-induced latency spikes. The root cause ultimately traced back to a single Linux kernel parameter, underscoring how low-level system settings can silently distort performance measurements at scale.