.png){kind=icon}
.png){kind=icon}
Return
Extended efficiency and soft-fairness multiresource allocation in a cloud computing system
.png){kind=icon}
PRE
AI
DOI：10.1007/s00607-022-01138-6.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Efficiency and fairness are two essential objectives for multiresource allocations in shared cloud computing systems. Due to the different demands of different users and the different capacities of each resource, it is impossible for multiresource allocations to achieve absolute fairness and maximum efficiency simultaneously. In this paper, we generalize dominant resource fairness (DRF) and propose a new allocation mechanism, max-min efficiency DRF (MME-DRF), to achieve a tradeoff between fairness and efficiency. MME-DRF first fairly allocates some resources to ensure a lower bound of relative soft fairness among users. Then, MME-DRF allocates the remaining resources with the goal of maximizing the minimum resource utilization. MME-DRF can obtain a max-min resource utilization that directly reflects the overall resource utilization of the system. Rigorous proofs show that MME-DRF satisfies four desirable properties, e.g., the sharing incentive, soft fairness, Pareto efficiency and weighted envy freeness. In addition, we develop an algorithm for MME-DRF and evaluate it via simulations driven by examples and Google cluster traces. The simulation results show that MME-DRF guarantees soft fairness and significantly improves the resource utilization of the system.
Keywords:
Cloud computing
Multiresource allocation
DRF
Max-min efficient
Soft fairness