Gianfranco Nencioni

Does Traffic Consolidation Always Lead to Network Energy Savings

In this paper we study the behavior of a general optimization model for reducing the power consumption of core networks employing energy-aware network equipment. Specifically, we assess how the energy profiles of the devices affect the outcome of the optimization model and hence determine the general power saving policy. The computational analysis performed on several real topologies shows that the widespread traffic consolidation strategy does not always provide the best results. In fact, for devices presenting a cubic (convex in general) energy profile, the highest energy savings are achieved by spreading the traffic on the network.

Mixed Integer Non-Linear Programming models for Green Network Design

The paper focuses on Network Power Management in telecommunication infrastructures. Specifically, the paper describes four energy aware network design problems, with the related mathematical models, for reducing the power consumption of the current and future Internet. Each problem is based on a different characterization and power awareness of the network devices, leading to either Mixed Integer Linear Programming or Mixed Integer Non-Linear Programming models. We have assessed the effectiveness of the proposed approaches under different real core network topology scenarios by evaluating the impact of several network parameters. To the best of our knowledge, this is the first work that deeply investigates the behavior of a pool of diverse Network Power Management approaches, including the first Mixed Integer Non-Linear Programming model for the Power Aware Network Design with Bundled Links.

A branch-and-Benders-cut method for nonlinear power design in green wireless local area networks

We consider a problem arising in the design of green wireless local area networks. Decisions on powering-on a set of access points (APs), via the assignment of one power level (PL) to each opened AP, and decisions on the assignment of the user terminals (UTs) to the opened APs, have to be taken simultaneously. The PL assigned to an AP affects, in a nonlinear way, the capacity of the connections between the AP and the UTs that are assigned to it. The objective is to minimize the overall power consumption of the APs, which has two components: location/capacity dimensioning costs of the APs; assignment costs that depend on the total demands assigned to the APs. We develop a branch-and-Benders-cut (BBC) method where, in a non-standard fashion, the master problem includes the variables of the Benders subproblem, but relaxes their integrality. The BBC method has been tested on a large set of instances, and compared to a Benders decomposition algorithm on a subset of instances without assignment costs, where the two approaches can be compared. The computational results show the superiority of BBC in terms of solution quality, scalability and robustness.

Robust optimisation of green wireless LANs under rate uncertainty and user mobility

Abstract We present a robust optimisation approach to energy savings in wireless local area networks, that incorporates both link capacity fluctuations and user mobility under Bertsimas and Sims robust optimization paradigm. Preliminary computational results are discussed.

Network Power Management: Models and Heuristic Approaches

The paper describes and compares different approaches that can be used to design Network Power Management methods, with the aim of reducing the power consumption of telecommunication networks. The approaches are based on the solution of optimization problems that have, in general, a Mixed Integer NonLinear Programming (MINLP) formulation. Given that the problems are NP-hard, exact methods for finding optimal solutions can be used only for scenarios of limited size. In this framework, the paper proposes a heuristic for finding a suboptimal solution of the Power Aware Routing and Network Design (PARND) problem, which is one of the more general design problems in Network Power Management. The simulation study highlights the capability of the proposed heuristic to obtain solutions near the optimum and to outperform the other approaches in terms of energy savings, while satisfying the constraints of the traffic demands.

The impact of the access point power model on the energy-efficient management of infrastructured wireless LANs

The reduction of the energy footprint of large and mid-sized IEEE 802.11 access networks is gaining momentum. When operating at the network management level, the availability of an accurate power model of the APs becomes of paramount importance, because different detail levels have a non-negligible impact on the performance of the optimisation algorithms. The literature is plentiful of AP power models, and choosing the right one is not an easy task. In this paper we report the outcome of a thorough study on the impact that various inflections of the AP power model have when minimising the energy consumption of the infrastructure side of an enterprise wireless LAN. Our study, performed on several network scenarios and for various device energy profiles, reveals that simple one- and two-component models can provide excellent results in practically all cases. Conversely, employing accurate and detailed power models rarely offers substantial advantages in terms of power reduction, but, on the other hand, makes the solving algorithms much slower to execute.

A New Markov Model for Evaluating the ALR Dual-Threshold Policy

The paper proposes the analysis of the Dual- Threshold Policy, used by Adaptive Link Rate (ALR). In par- ticular, the paper firstly presents a new Markov model of the Dual-Threshold Policy. The new features of the model concern its ability to take into account the rate switching times and the buffer size of the Ethernet Network Interface Controller (NIC). Secondly, we have evaluated the performance of the Dual- Threshold Policy by means of the proposed model and of a new power consumption model of an Ethernet NIC implementing ALR. As performance parameter, we have considered the power savings, the packet loss probability, the mean packet delay, and the oscillation of link data rate. The results show that high rate switching times can make ALR unprofitable and that the current NICs buffer size is unsuitable for ALR. I. INTRODUCTION Energy consumption has become a key issue during the past few years. The Internet is rapidly becoming a major consumer of power with significant economic and environmental im- pacts. Since most of power consumption is due to the access networks, the key considerations to reduce power consumption of networking devices are that Ethernet links are mostly underutilized (1) and that idle and fully utilized Ethernet links almost consume the same amount of energy. In particular, the energy consumption of an Ethernet Network Interface Controller (NIC) is proportional to the link data rate, i.e. 10M, 100M, 1G. Adaptive Link Rate (ALR) is a method to reduce energy consumption adapting Ethernet link data rate to link utilization. ALR is currently a key component of the Energy Efficient Ethernet (EEE) task force (IEEE 802.3az) in IEEE 802.3 (2). The key issues of the ALR are to define a mechanism for quickly switching the link data rate (ALR Frame Handshake Mechanism) and to create a policy to control the link date rate switching so as to maximize power savings without sig- nificantly worsening the network performance (ALR Control Policy). The first issue has been addressed by IEEE 802.3az, propos- ing a new mechanism called Rapid PHY Selection (RPS). RPS is based on a MAC frame handshake to initiate a link data rate switch. Following the handshake, the link would resynchronize at the new link rate. The sum of the MAC handshake time and the resynchronization time gives the rate switching time. During the rate switching period the link is inactive and no packets can be transmitted on it. Since during this period new packets can arrive at the Ethernet NIC, this can lead to buffer overflows and consequently to packet losses. Regarding the ALR Control Policies, they are outside the scope of IEEE 802.3az and remain a fertile area for research and evolution. In the last few years, several Control Polices have been already proposed ((3), (4), and (5)). In (3), the authors proposed the Dual-Threshold Policy, which is based on output buffer occupancy, or queue length, threshold crossing. In this policy, there are two thresholds used to induce hysteresis into the system and to prevent a trivial oscillation between rates. The policy works as follows. If the output buffer queue length in a NIC exceeds the higher threshold, then the link rate must be transitioned to high. If the output queue decreases below the lower threshold, then the link rate can be reduced to low. The main shortcoming of the Dual-Threshold Policy is that the NIC may oscillate between data rates given smooth traffic input or bursty traffic, with long duration bursts causing increased response time and variability.

SCORE: Exploiting Global Broadcasts to Create Offline Personal Channels for On-Demand Access

The last 5 years have seen a dramatic shift in media distribution. For decades, TV and radio were solely provisioned using push-based broadcast technologies, forcing people to adhere to fixed schedules. The introduction of catch-up services, however, has now augmented such delivery with online pull-based alternatives. Typically, these allow users to fetch content for a limited period after initial broadcast, allowing users flexibility in accessing content. Whereas previous work has investigated both of these technologies, this paper explores and contrasts them, focusing on the network consequences of moving towards this multifaceted delivery model. Using traces from nearly 6 million users of BBC iPlayer, one of the largest catch-up TV services, we study this shift from push- to pull-based access. We propose a novel technique for unifying both push- and pull-based delivery: the Speculative Content Offloading and Recording Engine (SCORE). SCORE operates as a set-top box, which interacts with both broadcast push and online pull services. Whenever users wish to access media, it automatically switches between these distribution mechanisms in an attempt to optimize energy efficiency and network resource utilization. SCORE also can predict user viewing patterns, automatically recording certain shows from the broadcast interface. Evaluations using our BBC iPlayer traces show that, based on parameter settings, an oracle with complete knowledge of user consumption can save nearly 77% of the energy, and over 90% of the peak bandwidth, of pure IP streaming. Optimizing for energy consumption, SCORE can recover nearly half of both traffic and energy savings.

Managed Dependability in Interacting Systems

A digital ICT infrastructure must be considered as a system of systems in itself, but also in interaction with other critical infrastructures such as water distributions, transportation (e.g. Intelligent Transport Systems) and Smart Power Grid control. These systems are characterised by self-organisation, autonomous sub-systems, continuous evolution, scalability and sustainability, providing both economic and social value. Services delivered involve a chain of stakeholders that share the responsibility, providing robust and secure services with stable and good performance. One crucial challenge for the different operation/control centres of the stakeholders is to manage dependability during normal operation, which may be characterised by many failures of minor consequence. In seeking to optimise the utilisation of the available resources with respect to dependability, new functionality is added with the intension to help assist in obtaining situational awareness, and for some parts enable autonomous operation. This new functionality adds complexity, such that the complexity of the (sub)systems and their operation will increase. As a consequence of adding a complex system to handle complexity, the frequency and severity of the consequences of such events may increase. Furthermore, as a side-effect of this, the preparedness will be reduced for restoration of services after a major event (that might involves several stakeholders), such as common software breakdown, security attacks, or natural disaster. This chapter addresses the dependability challenges related to the above-mentioned system changes. It is important to understand how adding complexity to handle complexity will influence the risks, both with respect to the consequences and the probabilities. In order to increase insight, a dependability modelling approach is taken, where the goal is to combine and extend the existing modelling approaches in a novel way. The objective is to quantify different strategies for management of dependability in interacting systems. Two comprehensive system examples are used to illustrate the approach. A software-defined networking example addresses the effect of moving control functionality from being distributed and embedded with the primary function, to be separated and (virtually) centralised. To demonstrate and discuss the consequences of adding more functionality both in the distributed entities serving the primary function, and centralised in the control centre, a Smart Grid system example is studied.

Benders Decomposition for a Location-Design Problem in Green Wireless Local Area Networks

Abstract We consider a problem arising in the design of green (or energy-saving) wireless local area networks (GWLANs). Decisions on both location and capacity dimensioning must be taken simultaneously. We model the problem as an integer program with nonlinear constraints and derive valid inequalities. We handle the nonlinearity of the formulation by developing a Benders decomposition algorithm. We propose various ways to improve the Benders master problem and the feasibility cuts.