Fair and Efficient Multi-Resource Allocation for Cloud Computing

TL;DR

This paper introduces a new fairness mechanism for multi-resource allocation in cloud computing that improves social welfare and resource utilization over existing methods, supported by theoretical analysis and experimental validation.

Contribution

It proposes a novel approximation ratio measure and generalizes the DRF mechanism to achieve better fairness and efficiency guarantees.

Findings

New mechanism outperforms classic DRF in social welfare and utilization

Theoretical analysis shows improved approximation ratios

Experimental results confirm effectiveness on real-world data

Abstract

We study the problem of allocating multiple types of resources to agents with Leontief preferences. The classic Dominant Resource Fairness (DRF) mechanism satisfies several desired fairness and incentive properties, but is known to have poor performance in terms of social welfare approximation ratio. In this work, we propose a new approximation ratio measure, called \emph{\fratio}, which is defined as the worst-case ratio between the optimal social welfare (resp. utilization) among all \emph{fair} allocations and that by the mechanism, allowing us to break the lower bound barrier under the classic approximation ratio. We then generalize DRF and present several new mechanisms with two and multiple types of resources that satisfy the same set of properties as DRF but with better social welfare and utilization guarantees under the new benchmark. We also demonstrate the effectiveness of these mechanisms through experiments on both synthetic and real-world datasets.