GMA: A Pareto Optimal Distributed Resource-Allocation Algorithm

TL;DR

This paper introduces GMA, a distributed algorithm for resource allocation in networks that guarantees Pareto optimality and prevents over-allocation, with proven theoretical properties and practical simulation validation.

Contribution

The paper presents GMA, a novel distributed algorithm for resource allocation that is Pareto optimal, scalable, and guarantees no resource over-commitment in network graphs.

Findings

GMA computes Pareto optimal allocations.

GMA guarantees no over-allocation of resources.

Simulation results show practical applicability of GMA.

Abstract

To address the rising demand for strong packet delivery guarantees in networking, we study a novel way to perform graph resource allocation. We first introduce allocation graphs, in which nodes can independently set local resource limits based on physical constraints or policy decisions. In this scenario we formalize the distributed path-allocation (PAdist) problem, which consists in allocating resources to paths considering only local on-path information -- importantly, not knowing which other paths could have an allocation -- while at the same time achieving the global property of never exceeding available resources. Our core contribution, the global myopic allocation (GMA) algorithm, is a solution to this problem. We prove that GMA can compute unconditional allocations for all paths on a graph, while never over-allocating resources. Further, we prove that GMA is Pareto optimal with respect to the allocation size, and it has linear complexity in the input size. Finally, we show with simulations that this theoretical result could be indeed applied to practical scenarios, as the resulting path allocations are large enough to fit the requirements of practically relevant applications.