Detection, Understanding, and Prevention of Traceroute Measurement Artifacts

📝 Abstract

Traceroute is widely used: from the diagnosis of network problems to the assemblage of internet maps. Unfortu- nately, there are a number of problems with traceroute methodology, which lead to the inference of erroneous routes. This paper studies particular structures arising in nearly all traceroute measurements. We characterize them as “loops”, “cycles”, and “diamonds”. We iden- tify load balancing as a possible cause for the appear- ance of false loops, cycles and diamonds, i.e., artifacts that do not represent the internet topology. We pro- vide a new publicly-available traceroute, called Paris traceroute, which, by controlling the packet header con- tents, provides a truer picture of the actual routes that packets follow. We performed measurements, from the perspective of a single source tracing towards multiple destinations, and Paris traceroute allowed us to show that many of the particular structures we observe are indeed traceroute measurement artifacts.

💡 Analysis

Traceroute is widely used: from the diagnosis of network problems to the assemblage of internet maps. Unfortu- nately, there are a number of problems with traceroute methodology, which lead to the inference of erroneous routes. This paper studies particular structures arising in nearly all traceroute measurements. We characterize them as “loops”, “cycles”, and “diamonds”. We iden- tify load balancing as a possible cause for the appear- ance of false loops, cycles and diamonds, i.e., artifacts that do not represent the internet topology. We pro- vide a new publicly-available traceroute, called Paris traceroute, which, by controlling the packet header con- tents, provides a truer picture of the actual routes that packets follow. We performed measurements, from the perspective of a single source tracing towards multiple destinations, and Paris traceroute allowed us to show that many of the particular structures we observe are indeed traceroute measurement artifacts.

📄 Content

arXiv:0904.2733v1 [cs.NI] 17 Apr 2009 Detection, Understanding, and Prevention of Traceroute Measurement Artifacts Fabien Viger a Brice Augustin a Xavier Cuvellier a Cl´emence Magnien a Matthieu Latapy a,∗Timur Friedman a Renata Teixeira a aUniversit´e Pierre et Marie Curie – CNRS, Laboratoire LIP6 Abstract Traceroute is widely used, from the diagnosis of network problems to the assemblage of internet maps. Unfortunately, there are a number of problems with traceroute methodology, which lead to the inference of erroneous routes. This paper studies particular structures arising in nearly all traceroute measurements. We characterize them as “loops”, “cycles”, and “diamonds”. We identify load balancing as a possible cause for the appearance of false loops, cycles, and diamonds, i.e., artifacts that do not represent the internet topology. We provide a new publicly-available traceroute, called Paris traceroute, which, by controlling the packet header contents, provides a truer picture of the actual routes that packets follow. We performed measurements, from the perspective of a single source tracing towards multiple destinations, and Paris traceroute allowed us to show that many of the particular structures we ob- serve are indeed traceroute measurement artifacts. Key words: traceroute, network topology measurement, measurement artifact, load balancing 1 Introduction Jacobson’s traceroute [1] is one of the most widely used network measurement tools. It reports an IP address for each network-layer device along the path from a source to a destination host in an IP network. Network operators and ∗Corresponding author. Address: LIP6 – 104 avenue du Pr´esident Kennedy – 75016 Paris – France Tel: +33 (0)1 44 27 87 84, Fax: +33 (0)1 44 27 74 95 Email address: Matthieu.Latapy@lip6.fr (Matthieu Latapy). Preprint submitted to Elsevier researchers rely on traceroute to diagnose network problems and to infer prop- erties of IP networks, such as the topology of the internet. This has led to an impressive amount of work in recent years [2,3,4,5,6,7,8], in which traceroute measurements play a central role. Some authors have noticed that traceroute suffers from deficiencies that lead to the inference of inaccurate routes, in particular in the presence of load balancing routers [3,4,9]. However, no systematic study of these deficiencies has been undertaken. Therefore, people dealing with traceroute measurements currently have no choice but to interpret surprising features in traceroute mea- surements as either characteristics of the routing or of the network’s topology. This is supported by the common assumptions that these deficiencies have a very limited impact, and that, in any case, nothing can be done to avoid them. The core contribution of this paper, which is a longer version of our earlier work [10], is to show that both of these assumptions are false. We show that the wide presence of load-balancing routers in the internet induces a variety of artifacts in traceroute measurements, and we provide a rigorous approach to both quantify and avoid many of them. More precisely, we focus on three particular structures often encountered in traceroute measurements, which we categorize as “loops”, “cycles”, and “di- amonds”. Using measurements from a single source tracing towards multiple destinations, we show that many instances of these structures are actually measurement artifacts resulting from load-balancing routers. We provide a new traceroute, called Paris traceroute, 1 which controls packet header con- tents to largely limit the effects of load balancing, and thus obtain a more precise picture of the actual routes. We show that many of the observed struc- tures disappear when one uses Paris traceroute. Finally, we explain most other instances using additional information provided by Paris traceroute, and sug- gest possible causes for the remaining ones. Throughout this paper, we use data obtained by tracing routes from one par- ticular source to illustrate our results (see Sec. 3). From the outset, we insist on the fact that this data is not meant to be statistically representative of what can be observed on the internet in general: it serves as an illustration only, and the quantities reported may differ significantly from what would be observed from other sources. Obtaining a representative view of the average behavior of the traceroute tool would clearly be of interest, but is out of the scope of this paper: we focus here on the identification of traceroute artifacts, their rigorous interpretation, and their suppression using Paris traceroute. This paper is structured as follows. Sec. 2 describes the classic traceroute 1 Paris traceroute is free, open-source software, available from http://www.paris-traceroute.net/ . 2 tool, its deficiencies, and the new tool we built to circumvent these deficien- cies, Paris traceroute. Sec. 3 describes our methodological framework. Sec. 4 categorizes the particular structures that we encounter

This content is AI-processed based on ArXiv data.