Using PeeringDB to Understand the Internet Peering Ecosystem

By: 
A. Lodhi, N. Larson, A. Dhamdhere, C. Dovrolis, K. Claffy
Appears in: 
CCR April 2014

In this study we mine one of the few sources of public data available about the interdomain peering ecosytem: PeeringDB [1], an online database where participating networks contribute information about their peering policies, traffic volumes and presence at various geographic locations. Although established to support the practical needs of operators, this data also provides a valuable source of information to researchers. Using BGP data to cross-validate three years of PeeringDB snapshots, we find that PeeringDB membership is reasonably representative of the Internet’s transit, content, and access providers in terms of business types and geography of participants, and PeeringDB data is generally up-to-date. We find strong correlations among different measures of network size - BGP-advertised address space, PeeringDB-reported traffic volume and presence at peering facilities, and between these size measures and advertised peering policies.

Public Review By: 
Renata Teixeira

Despite many years of research, the Internet structure and dynamics are still not fully understood. At the Autonomous System (AS) level, one important challenge is to infer the peering relationships between networks. This paper analyzes data from PeeringDB, which is a database where network operators publish information about their peering relationships together with other attributes such as traffic volumes and geographic location. The paper studies three years of PeeringDB data to first evaluate the extent of participation of ASes in PeeringDB and the attributes of these ASes declared in PeeringDB compared with these attributes inferred from BGP data. Then, the paper analyzes the changes in advertised peering policies over time. Reviewers were in general positive that this paper should be useful to the community. PeeringDB is an under-explored source of data that can help complement and validate inferences done using other sources of data (e.g., BGP). This paper clearly explains how this dataset can be useful. In particular, PeeringDB can be directly used to validate AS business types and can help infer connectivity between ASes. The reviewers did express concerns that the paper failed to discuss the limitations and inaccuracies of PeeringDB. The authors have added a discussion on the limitations of PeeringDB in the final version of this paper that helps point out the biases of the data. Still it is important that researchers who may want to use this dataset be aware that given the information in PeeringDB is self-reported we cannot take PeeringDB as ground truth without further analysis. One reviewer would have also liked to see a more in-depth analysis of the data than what is presented in the paper. In particular, PeeringDB data could be used to gain more insights into the changing nature of the industry. This study remains as a future work.