A School of Continuous Improvement paper deploys a fair-share job-scheduling algorithm, adapted from forty-year-old multi-user computing practice, on a suburban community makerspace's single 3D printer, finding it redistributes queue wait across member types more than it reduces it overall.
A twelve-week makerspace deployment lifts a fairness index into cluster-computing range, while the wait it distributes barely shrinks
For forty years, computing centres have used one trick to stop a heavy user monopolising a shared machine: track what they have already had, and make them wait longer for more. Slop University has now put that trick on a single 3D printer in a suburban community makerspace, and reports today what cluster-computing literature would have predicted, had anyone thought to ask it about a hobby workshop.
A paper published by Associate Professor Casimir Beng and Dr Mirela Hanke tracked a booking kiosk through a twelve-week term after its first-come-first-served queue was replaced with an exponentially-decayed fair-share formula, units changed but logic untouched from the multi-user computing literature. A weekly fairness index climbed into the range published cluster deployments report, while total time spent waiting barely moved: first-time and casual members waited substantially less, and the makerspace’s small population of heavy batch users waited substantially more — exactly the kind of scrutiny the School of Continuous Improvement expects any fairness mechanism it deploys to survive.
A fairness index and the queue you actually stand in are not the same measurement, and we would rather publish both than let one stand in for the other.
— Associate Professor Casimir Beng, Lead of the Adaptive Metrics Lab
The scheduler’s fourteen-day decay window lets a member’s recent printer use fade out gradually rather than resetting on a fixed cycle. Forty-seven members across the site’s full range of printing habits are covered by the twelve-week study; a follow-up ablation across three priority-weight settings found the weighting itself contributed little beyond the decay mechanism.
“It’s a shorter queue than anything we usually study, and that turned out to be the point,” said Dr Mirela Hanke, Postdoctoral Fellow and Deputy Convenor of the Living Dashboard. “A formula built for thousands of users still has to answer to the person standing in front of the printer.”
The full paper is available from the University’s research repository under an open licence, doi:10.5555/slop.ji65cu.