Impact of Multi-Access Links on the Internet Topology Modeling Mehmet Burak Akgun, Mehmet Hadi Gunes Department of Computer Science and Engineering University of Nevada, Reno Email: {makgun, mgunes}@cse.unr.edu Abstract—Comprehensive analyses that aim to better under- stand the topology of real world networks have been an impor- tant research challenge. Internet topology measurement studies provide samples of the underlying network at various levels. Although router-level Internet topology measurement systems target low level Internet infrastructure, they primarily focus on the Layer-3 connectivity and ignore the underlying multi-access links. In this paper, in addition to the thoroughly studied degree distribution, we analyze the subnet and interface distributions of major Internet topology datasets. We also investigate the impact of the higher granularity modeling at link level versus router level modeling. Our analysis establishes a foundation for the Layer-2 Internet topology generation and introduces the link layer characteristics into the network modeling. I. I NTRODUCTION Internet modeling mainly focuses on understanding local and global characteristics of the Internet at AS, PoP, or router level. This understanding is especially critical for newly developed protocols as the performance of protocols depend on the topological features of the underlying network [1], [2], [3]. If the synthetic topology used during simulation does not accurately represent the Internet, evaluation results will be misleading and the expected performance will not be observed when the system is deployed [4]. In general, Internet can be modeled as a set of inter- connected nodes based on some structural and/or functional relations. When we sample/generate topologies, many of the underlying relations may get broken or modified [5]. This may then result in a topology that does not resemble the original network for crucial characteristics [6], [7]. Researchers have studied various metrics that are found to be important in representing the reference graph and found that certain set of metrics are important for different networks. In this study, as we are interested in the link level Internet topology, we introduce interface distribution [8] and subnet distribution [9] as crucial metrics to closely model the un- derlying multi-access links in addition to the observed degree distribution. In order to capture the multi-access link characteristics, we analyzed the Internet topologies sampled by Ark [10], iPlane [11] and Cheleby [12] infrastructures. Our analysis reveals that the node degree, subnet size, and node interface distributions exhibit power-law patterns in the tail 1 . 1 A subnet-based topology generator is presented in [13]. In Section II, we evaluate current topology models in captur- ing the link-level characteristics, and point to inaccuracies with lack of multi-access links that reflect subnets. In Section III, we analyze large-scale characteristics of sample topologies. Finally, we discuss the implications of the multi-access links in network modeling and conclude in Section IV. II. ANALYSIS OF I NTERNET TOPOLOGY MODELS Subnets are the building blocks of the router-level Internet where multi-access links interconnect multiple systems over a single hop. Ideally multi-access links should be represented with a hyper edge that connects multiple nodes. Current router-level topology generators produce graphs composed of only point-to-point links. These topologies misrepresent the Internet as they ignore the multi-access links, which appear very frequently in the Internet backbone as measured in [14]. In modeling network topologies, we should consider the multi-access links [15] in addition to the observed degree distribution of the Internet [16]. In a topological perspective multi-access links have two major implications; pairwise one- hop connectivity and a collision domain between subnetwork devices. A. Subgraph Structures Fig. 1. Multi-Access vs. Point-to-Point Multi-access links may be mod- eled as a clique between attached nodes as shown in Figure 1 when the network model does not sup- port hyper edges. Note that this estimation is not perfect as neigh- boring subnets may incorrectly be assumed as a single subnet. In this section, we analyze whether current topology generators cap- ture the multi-access links by assessing the cliquness of the generated topologies. In order to compare with current generators, we sampled the subnet based Internet2 backbone topology 2 . Gathering ground truth information from configuration files of the routers, we had 440 nodes in 90 subnets where 60% of links were multi- access. In Figure 2-a, the Internet2 Subnet curve illustrates the subnet distribution of the Internet2 backbone. Distribution 2 Note that, even though router-level topology of other networks are publicly shared (e.g., GEANT,CENIC, and NLR), they do not contain the subnetwork information that is needed for a ground truth comparison.