Energy-Aware Traffic Engineering for Wired IP Networks
Luca G. Gianoli · PolyPublie (École Polytechnique de Montréal) · 2014
RESUME Meme si l'Internet est souvent considere comme un moyen formidable pour reduire l'impact des activites humaines sur l'environnement, sa consommation d'energie est en train de devenir un probleme en raison de la croissance exponentielle du trafic et de l'expansion rapide des infrastructures de communication dans le monde entier. En 2007, il a ete estime que les equipements de reseau (sans tenir compte de serveurs dans les centres de donnees) etaient responsables d'une consommation d'energie de 22 GW, alors qu'en 2010 la consommation annuelle des plus grands fournisseurs de services Internet (par exemple AT$T) a depasse 10 TWh par an. En raison de cette tendance alarmante, la reduction de la consommation d'energie dans les reseaux de telecommunication, et en particulier dans les reseaux IP, est recemment devenue une priorite. Une des strategies les plus prometteuses pour rendre plus vert l'Internet est le sleep-based energy-aware network management (SEANM), selon lequel la configuration de reseau est adaptee aux niveaux de trafic afin d'endormir tous les elements redondantes du reseau. Dans cette these nous developpons plusieurs approches centralisees de SEANM, afin d'optimiser la configuration de reseaux IP qui utilisent differents protocoles (OSPF or MPLS) ou transportent differents types de trafic (elastique or inelastique). Le choix d'adresser le probleme d'une maniere centralisee, avec une plate-forme de gestion unique qui est responsable de la configuration et de la surveillance de l'ensemble du reseau, est motivee par la necessite d'operateurs de maintenir en tout temps le controle complet sur le reseau. Visant a mettre en œuvre les approches proposees dans un environnement realiste du reseau, nous presentons aussi un nouveau cadre de gestion de reseau entierement configurable que nous avons appele JNetMan. JNetMan a ete exploite pour tester une version dynamique de la procedure SEANM developpee pour les reseaux utilisant OSPF.----------ABSTRACT Even if the Internet is commonly considered a formidable means to reduce the impact of human activities on the environment, its energy consumption is rapidly becoming an issue due to the exponential traffic growth and the rapid expansion of communication infrastructures worldwide. Estimated consumption of the network equipment, excluding servers in data centers, in 2007 was 22 GW, while in 2010 the yearly consumption of the largest Internet Service Providers, e.g., AT&T, exceeded 10 TWh per year. The growing energy trend has motivated the development of new strategies to reduce the consumption of telecommunication networks, with particular focus on IP networks. In addition to the development of a new generation of green network equipment, a second possible strategy to optimize the IP network consumption is represented by sleep-based energy-aware network management (SEANM), which aims at adapting the whole network power consumption to the traffic levels by optimizing the network configuration and putting to sleep the redundant network elements. Device sleeping represents the main potential source of saving because the consumption of current network devices is not proportional to the utilization level: so that, the overall network consumption is constantly close to maximum. In current IP networks, quality of service (QoS) and network resilience to failures are typically guaranteed by substantially over-dimensioning the whole network infrastructure: therefore, also during peak hours, it could be possible to put to sleep a non-negligible subset of redundant network devices. Due to the heterogeneity of current network technologies, in this thesis, we focus our efforts to develop centralized SEANM approaches for IP networks operated with different configurations and protocols. More precisely, we consider networks operated with different routing schemes, namely shortest path (OSPF), flow-based (MPLS) and take into account different types of traffic, i.e., elastic or inelastic. The centralized approach, with a single management platform responsible for configuring and monitoring the whole network, is motivated by the need of network operators to be constantly in control of the network dynamics. To fully guarantee network stability, we investigate the impact of SEANM on network reliability to failures and robustness to traffic variations. Ad hoc modeling techniques are integrated within the proposed SEANM frameworks to explicitly consider resilience and robustness as network constraints. Finally, to implement the proposed procedures in a realistic network environment, we propose a novel, fully configurable network management framework, called JNetMan. We use JNetMan to develop and test a dynamic version of the SEANM procedure for IP networks operated with shortest path routing protocols.