Approximation and online algorithms for selected network optimization problems
Year: 2013
Type: dissertation
Abstract: Optimizing energy efficiency is a fundamentally important issue in modern communication networks. In this thesis we study the objective of optimal energy usage on two general classes of wireless network optimization problems, namely network lifetime maximization and OFDMA scheduling. Orthogonal frequency division multiplexing (OFDM) has become an increasingly popular technology in the area of broadband communication. In one of the associated fundamental scheduling problems, data has to be sent from a base station to a collection of terminals using a given set of communication channels where, however, each channel can be assigned to at most one designated terminal. Aside from determining an appropriate feasible assignment, each channel needs to be allocated a communication rate such that - summing over all channels assigned to a single terminal - the total data rate satisfies at least the demand of that terminal. In order to guarantee a successful transmission of data at a certain rate, the transmission power on the corresponding channel has to be large enough. The objective is to minimize the total sum of energy requirements while satisfying the demands of each terminal. This and related problems (called OFDMA scheduling problems) have been intensively studied by the engineering sciences; however, we are the first to conduct a theoretical worst-case analysis of this practically relevant class of network optimization problems for which no lower nor upper bounds on the approximability were known before. While we will show that - already on a very restricted subclass - the problem of minimizing the required transmission power for given data rate demands can not be approximated by any (polynomial time computable) factor, we will also see that the dual objective of proportionally maximizing the data rates for a given total transmission power is approximable arbitrarily close to the optimum on this subclass. On general instances, OFDMA rate-maximization will be shown to be inapproximable by any factor better than 1/2. Complementarily, a simple matching-based 1/m-approximation algorithm will be derived (with m denoting the number of channels). Also a more involved convex programming approach is considered that (aside from its usefulness to derive better performance guarantees for special subclasses) can be used to compute 1/2-approximate half-integral solutions in the general~case. In the second part of this dissertation, we continue with our theoretical study of energy minimization in a more implicit context: Motivated by the behavior of battery depletion in wireless sensor networks we introduce and analyze a novel online model for general linear packing problems, applicable to a broad class of network lifetime maximization problems. Network lifetime maximization aims at sustaining a prespecified communication requirement as long as possible in a given communication network with limited power supplies. Though there exists a vast number of approximation approaches for a large variety of lifetime maximization problems, all previous formulations of network lifetime optimization assume perfect a priori knowledge about the status and future behavior of the energy supplies of each single network node. We relax this assumption by introducing an appropriate multiplicative uncertainty parameter which initially allows only a rough estimation of the energy capacities that becomes more and more accurate when exhausting the corresponding resources. The definition of the online model and the competitive online algorithms developed in this work are independent of the underlying problem structure as long as the corresponding optimization problem can be formulated in terms of a general linear packing program. An advantageous property of the proposed online algorithms is that they do not require any explicit linear programming formulation; hence, they are also suited when algorithmic efficiency is an issue. As the algorithms additionally are approximation factor preserving they can be used as a framework for algorithmically intractable problem classes by applying appropriate offline approximation algorithms. The proven competitive factors of the more sophisticated algorithms outperform naive online algorithms by at least a logarithmic factor with respect to the fixed uncertainty parameter. Complementarily, we show bounds on the competitiveness of arbitrary deterministic or randomized online algorithms in this online model.
Source: RWTH Publications (RWTH Aachen)
Authors: Marcel Ochel, Berthold Vöcking
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Keywords: network lifetime maximization, wireless sensor networks, energy efficiency, approximation algorithms, online algorithms, hardness of approximation, convex optimization, transmit power minimization, rate maximization, battery depletion, wireless networks, worst-case analysis, convex relaxation
Open Access status: green
OpenAlex ID: W2667210511
