Remote Peering Deployment and Its Implications for Internet Routing

10/06/2024

LACNIC collaborates with the development of applied research, supporting research groups from universities in the region. The Federal University of Rio Grande (FURG) is one of the organizations supported by LACNIC to develop its remote peering project, led by Professor Pedro Botelho Marcos.

In Botelho’s opinion, remote peering has grown notably in recent years, driven by Internet technology trends and the needs of the digital market.

This technical tool (PR) facilitated the growth of Internet eXchange Points (IXPs) and allowed smaller and geographically distant autonomous systems (AS) to connect more easily, the professor noted.

In this case, the research conducted with the support of the FRIDA program analyzed the effects of remote peering in four different aspects: growth of RP deployment over one and a half years; presence of route announcement mispractices (when networks prioritize the remote IXP over the local IXP), which are associated with routing anomalies; reliability of RP interfaces; and adoption of RP-related BGP communities (i.e. traffic engineering to remote peers).

The article seeks to shed light on the growth of PR at seven IXPs over the past year and explores some of the implications of the use of PR on the Internet.

The following key findings of the study are worth highlighting:

1. The growth of remote peering varies according to IXP characteristics. The study confirmed that remote interfaces have grown since 2021 in almost every IXP evaluated. The growth was directly related to how developed and prevalent the peering infrastructure was. Although growth was lower in well-established IXPs, remote peering development was predominant in more emergent infrastructures.

2. Undesirable trombone prefixes may be common, due to route announcements mispractices. ASes connected to multiple IXPs with a combination of local and remote connections may unintentionally occur in route announcement mispractices.

The analysis showed that more than 60 different ASes announced most specific prefixes on the remote IXP instead of prioritizing their local IXP connection. Over 37,000 prefixes were found with trombone paths, impairing peering performance.

3. Remote interfaces are more unstable than local interfaces. A concern about RP growth at IXPs is that networks using a shared port or being geographically distant would impose higher instability on their connection to the IXP RS, affecting other members. This is indeed the case in all analyzed IXPs. Remote interfaces seemed to be less reliable, presenting differences to local peers that reach up to 47.05% on a monthly analysis. Additionally, remote interfaces remain less time in ‘up’ state between failures, with local interfaces staying up to 3.7 times more active. Lastly, remote interfaces showed up to 3.5 times more state changes (from up to down) than their local peers.

4. Use of BGP communities to perform traffic engineering on remote peers is still not widespread. Many modern IXPs have begun to offer specific BGP communities to filter route export or perform some action (for example, prepend) to networks connected remotely. The analysis of these communities at three IXPs (PTT-SP, PTT-RJ, and PTT-CE) revealed that they are still negligible, representing less than 0.27% of all action communities observed at the IXP routes.

Botelho estimates that the findings helped characterize the implications of remote peering to Internet routing and its growth in different IXPs.

To improve the community’s understanding of the evolution and deployment of remote peering, we need to consider a larger number of IXPs.

We invite you to click on the following link to learn more about this study, which was supported by LACNIC’s FRIDA program.

The 2024 FRIDA Program call is open until June 28, more information here