Robust Energy-Aware Multi-period Traffic Engineering with Flow-based Routing Bernardetta Addis Department of Informatic Universit` a degli Studi di Torino Torino, Italy Email: addis@di.unito.it Antonio Capone, Giuliana Carello Department of Electronics and Information Politecnico di Milano Milan, Italy Email: carello@elet.polimi.it, capone@elet.polimi.it Luca G. Gianoli, Brunilde Sans` o D´ epartement de G´ enie ´ Electrique ´ Ecole Polytechnique de Montr´ eal Montr´ eal, Canada Email: luca-giovanni.gianoli@polymtl.ca, brunilde.sanso@polymtl.ca Abstract —A robust multi-period model is proposed to minimize the energy consumption of IP networks, while guaranteeing the satisfaction of uncertain traf- fic demands. Energy savings are achieved by putting into sleep mode cards and chassis. The study of the solution robustness shows that there is a trade-off between energy consumption and the solutions conser- vatism degree. The model allows this trade-off to be tuned by simply modifying a single parameter per link. The multi-period optimization is constrained by inter- period limitations necessary to guarantee network sta- bility. Both, exact and heuristic methods are proposed. Results show that up to 60% of the energy savings can be achieved for realistic test scenarios in networks operated with flow-based routing protocols (i.e. MPLS) and with a good level of robustness to traffic variations. I. Introduction Due to Internet rapid expansion, it is said that the ICT sector contribution is 2% (0.8 Gt CO2) of annual global greenhouse gas (GHG) emissions [1], and that in 2007 the Internet was responsible for 5.5% of the total energy consumption in the world [2]. Green Networking aims at optimizing telecommunication network energy consumption by working at different levels: development of i) new energy efficient network devices, ii) new method- ologies for power aware network design and iii) new energy management strategies [3]. The reader is referred to [4] for a discussion on different types of Green Networking proposals. In this paper we focus on IP network energy-aware management and we aim at limiting the energy-wise neg- ative effects due to bandwidth over provisioning, without reducing the QoS. In fact, although network utilization varies typically from 5% (night hours) to 50% (peak hours) [5], the network consumption remains practically constant because the energy consumed by network devices is almost independent of the traffic load [6]. A promising strategy is represented by energy-aware Traffic Engineering (TE), carried out assuming that un- used devices can be put to sleep. Energy-aware Traffic Engineering is strictly influenced by i) the routing protocol considered and by ii) the accuracy of the traffic estima- tions. In this paper we consider IP networks operated with Multi Protocol Label Switching (MPLS), that is, together with the Open Shortest Path First (OSPF) protocol, the most popular protocol adopted in the backbone IP networks. MPLS explicitly selects the route of each individual traffic demand, guaranteeing in this way a very flexible TE. Differently from previous work on IP networks energy savings, we specifically consider here the uncertainty of traffic estimations and the robustness of the network. We propose an offline method based on predicted values of traffic and on a robust optimization approach that assumes that traffic demands vary within a given uncertainty set [7]. We show that it is possible to obtain optimal solutions that satisfy network constraints even when the traffic demands do not exactly take the nominal values. We also show that there is a trade-off between robustness and energy consumption which can be tuned by just using a single parameter per link. The problem that we address in this paper is how to optimize network energy consumptions without affecting the network performance and the efficiency of network management mechanisms. For this purpose, we propose a multi-period optimization problem where we aim at minimizing the energy consumption of IP networks over a set of time intervals, while guaranteeing the satisfaction of the uncertain traffic demands. The value of each traffic demand can vary with a uniform distribution inside a symmetric interval centered on a nominal predicted value. A per-flow single path routing scheme is considered and energy savings are achieved by putting to sleep unused routers and links. Some inter periods constraints to limit the number of device switching on along the entire set of intervals are used to guarantee network stability and to preserve the expected device lifespan. An ILP formulation and a heuristic method based on the same ILP formulation are presented to solve the problem. The remainder of the paper is organized as follows. In Section II we review previous papers on green networking and point out the novelties of our work. In Section III-A we present the energy management strategy proposed, the system modelling assumptions and the ILP formulation,