PUBLISHED IN: PROCEEDINGS OF THE FUTURE NETWORK AND MOBILE SUMMIT 2010 CONFERENCE 1 On the incremental deployment of overlay information centric networks Konstantinos KATSAROS, Charilaos STAIS, George XYLOMENOS, George C. POLYZOS Athens University of Economics and Business, Patision 76, Athens, 104 34, Greece Tel: + 30 210 8203693, Fax: + 30 210 8203686 Email: ntinos@aueb.gr, stais@aueb.gr, xgeorge@aueb.gr, polyzos@aueb.gr Abstract— It has long been realized that the proliferation of information-centric applications and services must be reflected in a corresponding shift of the underlying Internet architecture. Even though users increasingly focus on the desired information, the underlying network still focuses on the endpoints provid- ing/consuming this information and in many cases this mismatch has resulted in an inefficient utilization of network resources, as demonstrated by peer-to-peer (P2P) and file sharing applications. In view of this situation, many research projects have focused on the investigation of alternative networking models centered around information. However, less attention has been paid to the transition process from the current end host centric model to an information centric one. In this paper, we propose an overlay multicast-enabled, publish-subscribe architecture and focus on its gradual deployment both inside administrative domain boundaries as well as across the Internet. Our simulation results demonstrate the benefits for individual network operators as they gradually adopt our new networking model, and shed further light on the extent of deployment required within an administrative domain in order for our approach to perform optimally. Index Terms— Content centric, multicast, publish-subscribe, content distribution. I. I NTRODUCTION The Internet protocols were originally designed to exchange traffic between pairs of communicating end hosts, following the prevailing communication patterns of previous networks. Communication patterns have evolved since then, and Internet use has shifted towards information-centric services and ap- plications, such as content delivery networks (CDNs), cloud computing services and peer-to-peer (P2P) file sharing. In these services, in sharp contrast to the underlying Internet model, the focus is on the information itself, rather than on the end hosts producing or consuming it. Hence, these services are implemented as overlays on top of the information-agnostic network substrate [1]. While the development of overlays has in many cases yielded important benefits to end users, it has also reviled the problems incurred by the mismatch between the prevailing communication patterns and the underlying network model. This is especially evident in the case of P2P file-sharing applications, in which end users benefit in terms of perceived download times and content providers decrease their resource requirements. However, the use of the (inappropriate) end-to- end communication model makes these applications plague the Internet with redundant unicast transmissions, e.g. many nearby nodes independently downloading the same data from a faraway node [2]. In view of this situation, in our previous work [3] we proposed the deployment of a router assisted overlay multicast (RAOM) architecture for information dissemination, based on the Scribe overlay multicast [4] and the Pastry overlay routing [5] schemes. The publish/subscribe nature of RAOM decouples information producers from consumers, thus break- ing the current end-to-end model. The deployment of overlay functionality inside the network instead of at the edges, as proposed by RAOM, leads to substantially improved multicast tree properties and reduced signaling overhead [3]. In this paper we focus on the transition to an ubiquitous RAOM deployment, by investigating the properties of the resulting overlay multicast trees in two directions. First we examine the benefits that would motivate individual network operators to invest on the RAOM provision, by comparing their perceived performance against operators that do not engage in RAOM. Second, we investigate the extent of the investment required inside domain boundaries in order to provide optimal performance, by studying the resulting tree properties for various deployment density degrees. The remainder of this paper is organized as follows. In Sec- tion 2 we present the RAOM architecture, while in Section 3 we discuss its incremental deployment. Section 4 first provides a thorough description of the simulation environment and then presents and discusses the results obtained. We conclude in Section 5. II. THE RAOM ARCHITECTURE The unsuitability of the Internet for information centric applications is evident in its lack of a multicast facility: multicast is inherently information-centric, it decouples the producers and consumers of information, and it promotes the efficient use of network resources. However, while IP multicast has been available for more than a decade, it has not been widely adopted for various reasons. IP multicast routing does not scale well: unlike unicast addresses that can be easily aggregated, nearly identical multicast addresses refer to completely different member sets, therefore routers must allocate memory and perform signaling separately for each group. There are also no gains to be made by supporting multicast, unless if all routers support it, therefore there are no incentives for individual routers to start doing so. Finally,