Internet reliability with realistic peering

Peering between network providers is modeled as a set of interconnections at hub nodes. The paper employs probabilistic methods to determine the likelihood of a path between nodes, under a variety of simulated disruptions or failures of parts of a network. Reliability is measured as the probability of withstanding failure of network components such as nodes or linkages. A reliability envelope is devised in order to define a range of impactsöwith the recognition that the damage to subsets of the network can vary from a relatively benign effect to a much more critical or lethal impact. Thus, we describe a range of scenarios from best to worst, within the same level of infrastructure loss. With the use of simulation data for the United States, results show the critical role of major hubs in sustaining communication among cities. In the paper we highlight especially reliable nodes (for example, Chicago and Atlanta) and those that are most susceptible (for example, Birmingham, Alabama, and Nashville, Tennessee). Failure in major hub cities such as Chicago, Dallas, and San Francisco would significantly influence adjacent areas and could cause severe degradation for the entire network with a loss of resiliency to below tolerable levels. An important finding is that various patterns in the reliability envelope can be influenced by factors such as network structures, hub structures, and geographical locations. DOI:10.1068/b3187 (1) IBPs are often referred to as `global Internet service providers' because they mainly transit huge amounts of Internet traffic at a national level (Huston, 1999). (2) Bandwidth is commonly defined as the information-carrying capacity of a link as measured in bits per second for transferring traffic. Massive investment for the growth in backbone capacity is among the most prominent trends in Internet development (Huston, 1999; Malecki, 2002). Center in NewYork highlights the issue of how vulnerable the Internet is to unexpected damage. The damage in New York City gave rise to short-term failures of statewide public online services for parts of the adjacent region. Further, it led to temporal network degradation for broader telecommunication services owing to the local loss of fiber optic lines, but at the same time the network exhibited remarkable resilience (see Grubesic et al, 2003; NYSDPS, 2002). In another example, a train crash in Baltimore's Howard Street tunnel in July 2001 caused the loss of fiber cables connecting two ISPs. Internet traffic in Washington ^ Baltimore became so congested that it had a significant impact on telecommunication quality on the East Coast, and this spread as far away as Seattle and Los Angeles. This incident also stressed the importance of peering arrangements with sufficient capacity on linkages among ISPs to make the Internet more resilient to disruption (Broadband Week 2001). The main purpose of this paper is to examine the vulnerability of the Internet in terms of pairs of major city nodes in the United States by taking into account peering arrangements in select hubs. In particular, in this paper we introduce the concept of reliability as a measurement to evaluate the vulnerability of the Internet as well as the geographical significance of selected network nodes. This paper is organized as follows. The following section explores issues relevant to an examination of the vulnerability of the Internet and to related analyses shown in previous research. Section 3 provides a brief explanation of reliability measurement as well as the characteristics of Internet hubs. The process of our model is demonstrated with the use of empirical analyses based on select IBPs in the United States. Concluding remarks are given in the final section. 2 Literature To our knowledge, the idea of a survivable network was initially addressed for various network configurations by Baran (1964). Since that time various approaches and measures have been proposed to either identify or assess network weaknesses (Ball et al, 1989; Coburn et al, 1994; Curley and Sha, 1982; Shake et al, 1999). In recent years the stability and tolerance of complex topologies such as scale-free and exponential networks with respect to random failures have been explored using probabilistic approaches. The results show a remarkable resiliency of large-size networks against random disruption, but also stress that current telecommunication systems could be vulnerable to failures at vital hubs (Albert and Barabäsi, 2002; Albert et al, 2000; Callaway et al, 2000; Yook et al, 2001). Work on the vulnerability of the Internet has also been found recently in geography. Geographers working on Internet backbones created a ranking for US cities in terms of measurements such as connectivity and graph theory (Malecki and Gorman, 2001; Moss and Townsend, 2000). The differences of city accessibilities and topological structures of the US commercial Internet were also investigated through the use of network analysis (O'Kelly and Grubesic, 2002; Wheeler and O'Kelly, 1999). Other work measures the growth of the Internet by comparing bandwidths of network links for cities or metropolitan areas (Malecki, 2002). The previous research mainly focuses on the potential availability for each commercial Internet network. The common characteristics in this work is that it assumes that all backbone networks of the Internet operate normally without any failures of network components such as links and nodes. Potential availabilities of a network are generally expressed in the form of discrete scores such as the number of paths, accessibility indices, and total volumes of capacities such as bandwidth. Therefore, the performance of commercial networks can be evaluated by comparing their relative values. 326 M E O'Kelly, H Kim, C Kim