On quality of service optimization with discrete QoS options

TLDR

QoS management seeks to allocate system resources to maximize end‑user utility, previously assuming continuous QoS dimensions and concave utility functions. The paper introduces a QoS framework that relaxes these assumptions and formulates resource allocation as a utility‑maximization problem, proposing two near‑optimal algorithms. The framework supports discrete QoS operating points and arbitrary utility functions, and the algorithms guarantee bounded sub‑optimality compared to a dynamic‑programming optimal baseline. Empirical evaluation shows which algorithm is suitable for online real‑time deployment.

Abstract

We present a QoS management framework that enables us to quantitatively measure QoS, and to analytically plan and allocate resources. In this model, end users' quality preferences are considered when system resources are apportional across multiple applications such that the net utility that accrues to the end-users is maximized. We previously worked with continuous QoS dimensions, and assumed that the utility gained by improvements along a QoS dimension were always representable by concave functions. In this paper we relax both assumptions. One, we support discrete QoS operating points. Two, we make no assumptions about the concavity of the utility functions. Using these as the basis, we tackle the problem of maximizing system utility by allocating a single finite resource to satisfy the QoS requirements of multiple applications along multiple QoS dimensions. We present two near-optimal algorithms to solve this problem. The first yields an allocation within a known bounded distance from the optimal solution, and the second yields an allocation whose distance from the optimal solution can be explicitly controlled by the QoS manager, We compare the run-times of these near-optimal algorithms and their solution quality relative to the optimal allocation, which in turn is computed using dynamic programming. These detailed evaluations provide practical insight into which of these algorithms can be used online in real-time systems.