Multi-resource Fair Allocation with Bounded Number of Tasks in Cloud Computing Systems

摘要

Dominant resource fairness (DRF) is a popular mechanism for multi-resourceallocation in cloud computing systems. In this paper, we consider a problem ofmulti-resource fair allocation with bounded number of tasks. Firstly, wepropose the lexicographically max-min normalized share (LMMNS) fair allocationmechanism, which is a natural generalization of DRF, and design a non-trivialoptimal algorithm to find a LMMNS fair allocation, whose running time is linearin the number of users. Secondly, we prove that LMMNS satisfies envy-freeness(EF) and group strategy-proofness (GSP), and analysis the approximation ratiosof LMMNS, by exploiting the properties of the optimal solution. Thirdly, wepropose a modified version of LMMNS, which is the second mechanism satisfyingsharing incentive, EF, and GSP. Finally, we have implemented LMMNS, and showthat it has a good average-case performance, especially when the number ofresources is 2.