Skip to main content
Illustration for The Half-Life of Usage: Fair-Share Decay and Post-Fieldwork Bursts on a Shared University Batch Cluster

The Half-Life of Usage

Fair-Share Decay and Post-Fieldwork Bursts on a Shared University Batch Cluster

A paper from the School of Underlying Conditions replays twenty-six months of job records from the University's 4,608-core shared batch cluster under seven fair-share decay half-lives, from one day to eight weeks, and finds that mean wait moves by under ten per cent while several hours of tail wait pass between the eleven accounts that submit in post-fieldwork bursts and the fifty that submit steadily. Under long half-lives the bursty accounts' worst waits fall in the tail of a burst, among short re-runs, and a closed-form model of decayed usage places the knee near the median burst length times ln 2, about thirteen days.

Authors Vigdis Emenike , Mirela Hanke
School School of Underlying Conditions
Output type Research paper
Published
DOI 10.5555/slop.9k2upu
Pages 6
Version 1.0
Licence CC BY 4.0

Cites

Cite as

@misc{slop_9k2upu,
  author       = {Vigdis Emenike and Mirela Hanke},
  title        = {The Half-Life of Usage: Fair-Share Decay and Post-Fieldwork Bursts on a Shared University Batch Cluster},
  year         = {2026},
  publisher    = {Slop University},
  doi          = {10.5555/slop.9k2upu},
  url          = {https://slop.university/outputs/slop-paper-how-the-fair-share-9k2upu/},
  version      = {1.0},
  note         = {Research paper},
}