Skip to main content
A household wifi's fairness mode keeps the same device on top

A household wifi's fairness mode keeps the same device on top

A new Slop University paper instruments 34 household routers for sixteen weeks, finding a router's AI-optimised fairness mode never closes the bandwidth gap favouring the oldest-connected device, and that its own log records the gap as resolved regardless.

Instrumenting thirty-four routers for sixteen weeks finds an AI-optimised fair-share mode never moves the bandwidth gap it logs as resolved

For sixteen weeks, a small logger spliced into the networks of thirty-four households watched a bandwidth argument nobody in any of them had actually settled. Every router in the study already claimed to have settled it.

The School of Continuous Improvement has built a research programme out of pointing its evaluation instruments at whatever a household already assumes is working, and a wifi router, its newest paper finds, qualifies as readily as anything in a workplace. Across six commercial router families, each marketed as running an AI-optimised mode that redistributes bandwidth fairly during congestion, the device holding the oldest continuous connection kept a mean 1.9 times its fair share of the link whenever the household’s wifi was under real strain.

A log that says “resolved” is not the same claim as a log that says “equal.” A household deserves to know which one its router is actually making.

— Associate Professor Casimir Beng, Lead of the Adaptive Metrics Lab, School of Continuous Improvement

A four-week trial forcing every router to reboot nightly, meant to reshuffle which device connected first, left the gap untouched. “The reboot was supposed to shuffle the deck,” said Dr Torun Ezeigwe, Senior Lecturer and Director of the Master of Applied Measurement. “It turns out some devices deal themselves back in before anyone else has cut the cards.” The fairness mode’s own admin-panel log, meanwhile, recorded the gap as resolved in three windows out of five, whether or not any device’s share had actually moved.

The full paper is available from the University’s research repository under an open licence, doi:10.5555/slop.mvztwy.