Two new Internet topology datasets

Tue, 09/15/2009 - 12:42 by Olivier Bonaventure • Categories:

The research community has developed many tools to try to map the topology of the Internet. As the Internet is a decentralised network composed of more thank 30000 different networks, it is very difficult (and perhaps impossible) to obtain a detailed map of the network. The most widely use to infer the router-level topology of the Internet is traceroute. Initially developed as a network debuging tool by Van Jacobson, traceroute allows to detect routers on the path between a source and a destination by sending packets with different time-to-live values. Among all traceroute users, CAIDA has collected a large dataset for its skitter and arkchipelago projects. These datasets provide useful information about the Internet topology. However, in such datasets, it is difficult associate the detected IP addresses to the corresponding routers. This alias resolution problem was analysed by several researchers, but there is no perfect solution.

In an ideal world, researchers could use SNMP to send queries to all routers to detect all their IP addresses and build a full map of the Internet based on these SNMP replies. Unfortunately, almost no ISP router replies to external SNMP queries and some operators consider remote SNMP queries as attacks that must be prosecuted. Fortunately, there is one little known alternative. In the early days of wide area IP Multicast with the MBONE, it was difficult to debug problems in the multicast topology without having information about the routers. During the development of the DVMRP protocol in the mrouted software, the mrinfo tool was introduced. mrinfo relies on an extension of the IGMP protocol described in http://www.potaroo.net/ietf/idref/draft-ietf-idmr-dvmrp-v3/
mrinfo can query a multicast router and receive the list of all IP addresses of this router with their status. This is a very interesting tool for researchers willing to map the Internet. Although the tool was introduced in mrouted, it is also supported by cisco and juniper routers as soon as IP Multicast is enabled on such routers. Furthermore, Jean-Jacques Pansiot has collected daily mrinfo reports from thousands of routers during the last four years. We will present a detailed paper showing the benefits of mrinfo to map network topologies, with an emphasis on interdomain links, will be presented at IMC2009 in November 2009.

Besides traceroute, researchers often use dumps form BGP routing tables to infer the AS-level topology and the peering links between ASes. When using a large number of dumps, it is possible to see lots of customer-provide links, but shared cost peering links are much more difficult to extract from BGP dumps. During IMC2009, Brice Augustin, Balachander Krishnamurthy and Walter Willinger will present an alternative approach that relies on targeted traceroutes sent via Internet Exchange Points to infer shared cost peering links and they have released their dataset.

By exploiting these two new sources of data, we can expect to have a more detailed map of the global Internet.