Network maps beyond connectivity

Zhiheng Wang, Cheng Jin, S. Jamin
2005 GLOBECOM '05. IEEE Global Telecommunications Conference, 2005.  
Knowing network topology is becoming increasingly important for a number of applications such as server placement [1] and traceback of DDoS attacks [2] . Recent works in modeling the Internet topology and constructing network maps have focused on the connectivity aspect. This paper describes the first study in incorporating connectivity, latency, and routing information all into a network map based on a large set of traceroute data. We highlight the common challenges in constructing such
more » ... maps and discuss our solutions. We evaluate our network map based on various Internet routing models proposed in the literature. The evaluation shows that, for those traceroute data that we are able to evaluate, at least 85% of computed hop-counts and latencies are within a factor of two of the actual values. Furthermore, we show that a flat routing model based on hop-count performs as well as more complicated policy routing models. Introduction There is a frequent need in network research to learn the topology of the Internet or other networks of interest. The knowledge of network topologies proves useful in two ways. First, studies to uncover properties of network topologies can be made based on real network topologies. There have been a number of studies [3, 4, 5, 6, 7 ] that try to understand and model the Internet topology based on actual data. Second, the actual network topological information can be used by a number of applications and services, such as CDN placement [1, 8] , and prevention of DDoS attacks [2, 9] . AS-level connectivity can usually be obtained from BGP routing table dumps [10, 6] . It is generally more difficult to obtain host-level connectivity since host-level routing tables are not readily available. There have been a number of studies [11, 12, 13 ] that focused on building host-level connectivity maps. Authors in [11] build network maps for visualization of network connectivity. Authors in [12] capture the transit portion of the Internet to study path inflation by policy routing, and authors in [13] focus on building complete connectivity maps of ISPs. All three mapping efforts produce only connectivity maps. Connectivity information is sufficient to describe AS-level topology since no other metrics such as latency or bandwidth can be meaningfully assigned between many large ASs. Such is not the case with host-level network maps. Host-level network topologies also contain information such as latency and bandwidth. Latency and bandwidth information are often highly useful for performance studies. For example, for studies on TCP's throughput [14] , in either simulations or real experiments, end-to-end latency determines how quickly a sender detects network congestion, and determines the throughput along with the bottleneck bandwidth. In simulations, end-to-end latencies and bottleneck bandwidths generally computed based on some routing model that computes paths among nodes inside a network. Presently, there are no studies on building a network map with more than connectivity information or on the best routing model to use for path computation. In this paper, we combine connectivity, latency, and routing information all into an interface-level network map based on Internet traceroute data. The interface-level network map differs from the host-level network map in that only IP addresses, not actual hosts, are represented in the map. We will not attempt to resolve interface aliases [12, 6] , i.e. a single router having multiple interfaces, since the network metrics we use for evaluations, hop-counts and latencies, are valid only on the IP interface level. We emphasize that we are not proposing a network topology generator in this paper; instead, we are interested in utilizing a given set of Internet measurements to generate the most comprehensive network map covered by the measurements. We hope that an accurate network map based on real measurements will be useful to applications and services that require topological information, and will provide the basis for a more complete model of Internet topology. Our contributions in this paper are as follows: • A detailed study on how to incorporate link latencies into topology construction. • A survey of routing models that can be used for path computations. • An evaluation of the network map showing that we can indeed compute hop-count and latency information with reasonable accuracy on this network map.
doi:10.1109/glocom.2005.1577668 dblp:conf/globecom/WangJJ05 fatcat:rys5c6uf45h3rcoygnu5edjnwy