Stateful DRF: Considering the Past in a Multi-Resource Allocation

摘要: The multi-resource allocation problem arises in different scenarios. Different mechanisms have been proposed to fairly divide multiple resources, most notably, Dominant Resource Fairness (DRF). Even though DRF satisfies several desirable properties, it considers fairness only the static setting. We propose Stateful (SDRF), an extension of that looks at past allocations and enforces long run while keeping fundamental properties DRF. prove SDRF is strategyproof, since users cannot manipulate system by misreporting their demands; incentivizes sharing, because no user better off if resources are equally partitioned; efficient, as can be improved without decreasing another. In SDRF, priorities change over time. To avoid recalculating every task scheduling decision, we also Live Tree, a data structure keeps elements with predictable time-varying ordered. implement on Mesos real cluster. Moreover, conduct large-scale simulations based Google cluster traces 30 million tasks one month. Results show reduces users' waiting time average. This improves fairness, increasing number completed for lower demands, negligible impact high-demand users.