Study of Internet autonomous system interconnectivity from BGP routing tables
Introduction
An autonomous system (AS) is a set of routers under a single technical administration, using one or various IGPs (Interior Gateway Protocol, i.e., RIP, OSPF, etc.) and common metrics to route packets inside the AS and using an EGP (Exterior Gateway Protocol, i.e., BGPv4) to route packets to other ASs. BGP comes from Border Gateway Protocol and is an inter-autonomous system routing protocol, [16]. BGPv4 is extensively used to connect ISPs (Internet Service Providers) and to interconnect enterprises to ISPs. ISPs usually are providers (provide connectivity) of other ISPs that at the same time are providers of smaller ISPs. In the periphery of the Internet there are end ISPs that usually give services to enterprises that are not ISPs. ISPs can be classified as transit ISPs when they offer transit of traffic, multihomed ISPs when they are connected to more than one ISP and do not offer transit of traffic, and stub ISPs when they are connected to other ISPs only. An ISP can have more than one AS number assigned and give services to other ISPs on large geographical areas. We will consider for simplicity that ISPs are single autonomous systems, and we will study topologies based on BGP interconnectivity. These assumptions are not far from reality since the Internet topology is based on inter-domain interconnectivity and routing policies handled by BGP.
The primary function of BGP is to exchange network reachability information with other BGP peers on neighbour ASs. ASs can apply different policies to the way they import/export routes when using BGP. Routing policies define how routing decisions are taken in the Internet. If we have two networks separated by two routers belonging to different ASs, the policy comes into play when one AS decides to announce networks (prefix routes) to the other AS. Import policies allow an AS to transform incoming route updates. For example, denying or allowing an update, assigning a local preference attribute to an incoming route depending on the AS origin, AS path, etc. ASs only send their best route to their neighbours. Export policies allow an AS to determine whether to send this route to a neighbour and, if it does, to send the route with or without hints such as the community attribute or the MED (Multi Exit Discriminator). A route is expressed as a prefix, i.e., a group of one or more networks. Routing policies are not applied to each prefix separately but to a group of prefixes defined at the AS.
BGP allows reachability information exchange among BGP peers of the same AS or of different ASs. This information will allow us to construct a graph of ASs connectivity, this is to say, the Internet topology at autonomous systems level. A BGP peer builds a routing table database consisting of the set of all feasible paths and the list of networks (prefixes) reachable through each feasible path or AS path. An example of a BGP table is shown in Table 1, where the symbol “>” expresses the “best path”. As BGP routers only send their best path to BGP peers, a BGP router will have a particular view of the Internet topology depending on where it is placed. In order to have a wider perspective of the Internet topology, it is necessary to study the union of several BGP tables of different places, as we do in this work. Oregon Route Views, [3], is a repository that saves every two hours the BGP tables of ASs connected to the BGP route repository. Oregon Route Views uses AS6447 and it is currently connected to 60 neighbours.
Based on the work of [9] and making use of the Oregon Route Views BGP table [3], many researchers have recently investigated the Internet topology, [5], [6], [7], [10], [14], [17]. These studies are relevant to the development of Internet topology generators, the deployment of content networks, the placing of web servers or in the development of inter-domain traffic engineering models, etc.
Authors of [8], [17] argue that tables taken from only one point such as the Oregon Route Views give a poor vision of the Internet and they propose to use more than one point of vantage. Authors from CAIDA, [13], propose to generate active probes to complete the AS interconnectivity. Bu and Towsley, [6], propose using the CCDF (complementary cumulative distribution function) to better fit the power-law behaviour described by Faloutsos et al. Finally, Lakhina et al, [15], describe sampling biases in IP topology measurements using traceroutes.
Here, we compare several BGP tables from different geographical sites and their union, using classical network topology measures: AS path distribution, clustering coefficient and degree distribution together with information about the number of ASs and the number of edges seen in the different perspectives. We will compare tables of different sizes and from ASs with different interconnectivity and we will join the repository of Route Views with these BGP tables. This work is part of the work published in [12].
Other authors have centred their study on the novel definition of complex networks. A complex network shows certain organization principles that are encoded in its topology. These works study three main topology models: the classical Erdös–Renyi model for random graphs, the small-world model motivated by small paths between two nodes and high clustering coefficients and the scale-free model that presents power-law degree distributions. Some of these works are [4], [18] or [20].
We will discuss the use of scale-free models to generate Internet topologies. In particular we study the application of scale-free models to the whole BGP table and to part of the BGP table. Using well-known heuristics proposed by Gao in [10], [11], we infer peering relationships between ASs and take away end customers and small ISPs from the BGP table as is done in [17]. In this way we can analyze the use of scale-free models applied to the core of the Internet and identify the kind of peering of any AS.
In Section 2 we define the metrics selected to compare the BGP tables. Section 3 is devoted to the explication of the methodology used. Section 4 shows the obtained results applied to the whole BGP table, data used in this section is dated on October/2002. Sections 5 shows the results applied to parts of the BGP table, data used in this section is dated on april/2003. Section 6 finalizes with the conclusions of the work.
Section snippets
Metrics
In this section we define the metrics selected to compare our sources of data. For that purpose we consider the AS level topology as the graph , where N is the number of vertices or ASs and E the number of edges or links that connect the vertices. We define the adjacency matrix A as a symmetric matrix of size NxN with components aij=1 if node i has an edge joining node j and 0 otherwise.
At each of the different BGP tables we will investigate the AS degree rank, the AS degree CCDF
Methodology
In this work, we use besides the data of [3] six public available BGP table [1]. In order to get the adjacency matrix, {aij}, we need to analyze BGP routing tables, which provide us with AS paths and links contained in them. It is important to note that BGP is a protocol of peering relationships and not of physical connections. For that reason the local view of an AS located in Europe could be (and is) different of an AS located in Asia and so on. This fact motivated us to investigate the
The whole BGP table
In this section we will investigate the AS connectivity. We analyze each of the BGP Tables and their union.
Splitting the BGP table in three regions
Recently, some works have questioned the application of the AB model to Internet topologies, see [8] or [20]. The major criticism made to the AB model is that the model does not take into account the dynamics of the BGP routing in the Internet and the business and geographic preferences. Some of these aspects could be studied identifying the end customers from the small provider ISPs and the core-transit ISPs as done in [17] and knowing whether ASs with low degree are connected with nodes with
Conclusion
In this paper, we compare BGP tables from different sites and of different sizes. We have chosen six complete BGP tables. We have obtained the adjacency matrix A of these tables and of the union of all the tables. Since the degree rank and the degree density function do not fit a power-law 1 very well, we have chosen the degree CCDF. We have shown that the degree CCDF follows a power-law Fd∝d−α that fits better than the degree density function and that the more complete the data sources are,
Acknowledgements
This work was supported by the Ministry of Education of Spain under grant CYCIT TIC-2001-SGR-00226 and by CIRIT under grant CIRIT 2001-SGR-00226 and NoE EuroNGI.
José M. Barceló received his engineering degree in Telecommunication Engineering in 1991 and his Ph.D. in 1998 from the Polytechnic University of Catalonia (UPC). He is full-time associate professor in the Computer Architecture Department of the UPC where he does research in the Traffic Management for Integrated Services Networks Group. During the past years, he has worked in several European projects (RACE and ACTS) related to ATM, access networks, TCP/IP and mobile IP.
References (20)
- Public route server and looking glass list”, Available from...
- RIPE Routing Information Service, Available from...
- Route Views “University of Oregon Route Views Project”. Available from...
- et al.
Statistical mechanics of complex networks
Reviews of Modern Physics
(2002) - A. Broido, K.C. Claffy, Internet expansion, refinement and churn, ETT January...
- T. Bu, D. Towsley, On distinguishing between Internet power-law topology generators, in: Proceedings of INFOCOM,...
- H. Chang, R. Govindan, S. Jamin, S.J. Shrenker, W. Willinger, On inferring AS-level connectivity from BGP routing...
- Q. Chen, H. Chang, R. Govindan, S. Jamin, S.J. Shenker, W. Willinger, The origin of power-laws in Internet topologies...
- M. Faloutsos, P. Faloutsos, C. Faloutsos, On power-law relationships of the Internet topology, in: Proceeding of ACM...
- et al.
On inferring Autonomous system relationships in the Internet
IEEE/ACM Transactions on Networking
(2001)
Cited by (19)
How to provide Quality of Service guarantees in the Internet? Lessons learnt from the airline and card-based payment sectors
2011, European Management JournalCitation Excerpt :Table 1 shows the distribution of the top one hundred ISPs according to their hierarchical level (Tier 1, 2, or 3) at three different periods (July 2005, January 2006 and July 2006). Several studies have shown that the Internet, and particularly the World Wide Web, exhibit a scale-free topology (Albert & Barabási, 2002; Barabási & Albert, 1999; Bu & Towsley, 2002; Faloutsos, Faloutsos, & Faloutsos, 1999; Medina, Lakhina, Matta, & Byers, 2001; Palmer & Steffen, 2000; Vásquez, Pastor-Satorras, & Vespignani, 2002) and small world features (Barceló, Nieto-Hipólito, & Garcia-Vidal, 2004; Jin & Bestavros, 2006). The originality of our article is to examine the graph of relations between ISPs through a hierarchical perspective.
Sensitivity metrics of complex network based on co-occurrence truth table: exemplified by a high-speed rail network
2023, Journal of Geographical SystemsA comprehensive survey on offload techniques and management in wireless access and core networks
2015, IEEE Communications Surveys and TutorialsResearch on relevance between k-core and clustering coefficient in complex network
2015, Tongxin Xuebao/Journal on CommunicationsAnalysis of fractal characteristic of Internet router and IPv6 level topology
2014, Dongbei Daxue Xuebao/Journal of Northeastern UniversityResearch on internet macroscopic topology structure
2013, Advanced Materials Research
José M. Barceló received his engineering degree in Telecommunication Engineering in 1991 and his Ph.D. in 1998 from the Polytechnic University of Catalonia (UPC). He is full-time associate professor in the Computer Architecture Department of the UPC where he does research in the Traffic Management for Integrated Services Networks Group. During the past years, he has worked in several European projects (RACE and ACTS) related to ATM, access networks, TCP/IP and mobile IP.
Juan I. Nieto-Hipólito received his M.Sc. in Telecommunication Networks form CICESE, Ensenada, México in 1994. He is associate professor in the Universidad Autónoma de Baja California (UABC) and he actually is in the Computer Architecture Department of the Technical University of Catalonia (UPC), where he is a PhD candidate with a grant of the Mexican government.
Jorge Garcı́a-Vidal obtained his Telecommunication Engineering degree (1988) and his Ph.D. degree in Telecommunication Engineering (1992) in Polytechnic University of Catalonia, UPC (Award to the Best Ph.D. in Telecommunication by UPC (award to the best Ph.D. by ANIEL and COIT). During 1992-93 he was research visitor at SIE department of University of Arizona, with a NATO fellowship. Since 2003 he is full professor at UPC. He has done research in performance evaluation of computer systems, medium access control protocols, wireless LANs, TCP/IP protocols implementation and in mobility in IP networks. He has participated in several EU funded research projects and in projects in collaboration with industry.
- 1
Belongs to the Autonomous University of Baja California (UABC), México. He is a PhD student at UPC, Spain, with a grant of the Mexican government.