Last year, the Internet Society performed a small study to explore the development and innovation happening at the edge of the Internet, focusing on enterprises, access networks, and data centers.
We hypothesized that the following changes in the usage and importance of these services could impact the relevance of the classic Internet:
- The decline in transit services, which allows traffic to escape one’s locality, and what makes the Internet a truly global network.
- The development and growth of regional or global ‘edge networks’, such as Cloud and Content Deliver Networks (CDNs).
- Changes in the access networks with the advent of fiber to the home (FTTH) and 5G.
The Three Types of Internet
To understand the effect of declining transit, we first performed a desktop study on how the evolution of edge services is taking place, what requirements drive this evolution, and whether there is a drive to different economic models.
From this, we observed three evolutionary forms of the Internet that co-exist and complement each other:
- A classic Internet, which is represented by a mesh of interconnected networks across which resources are spread.
- A hybrid Internet, which is characterized by a concentration of resources in content delivery networks (CDNs) and Cloud networks delivered close to the consumer through local nodes and caches.
- A telco Internet, where the resources are brought even closer to end users.
We envision that the hybrid Internet will serve most applications, that is, the data will be so close that current-day Internet CDN models can serve those applications and their users without fundamental architectural changes. In this model, a lot of traffic will flow locally but transit is still needed to fill cache infrastructure (if it’s not filled by the cache provider’s private infrastructure) or provide users access to the long tail of infrequently used web content and other services.
What Impact is this Evolution Having on Service and Content Providers?
We also assessed the dependencies between the three models of the Internet with a specific focus on classic and hybrid—we wanted to understand how the industry is evolving with the rise of the edge.
With the big caveat that this is all anecdotal (and can serve as a thesis for further research), we want to share what we learned based on a few interviews.
The ISP Perspective
From the perspective of Internet Service Providers (ISPs), the dependency on transit is ranging from 10% to 30% of the total traffic. This dependency is likely to drop since, for ISPs, a more commercially viable strategy is to reduce reliance on transit by expanding the peering network (either at Internet exchange points (IXPs), by private interconnects, or by connecting to another IXP). The dependencies become weaker due to increased concentration on the service and content part.
Most wanted content is served through a handful of CDNs and services increasingly move to Cloud, dominated by a few major providers. That means that most customer demands can be satisfied with only a few direct peering relationships, and the need for transit diminishes.
As far as essential services, the Domain Name System (DNS) and Network Time Protocol (NTP)—the service that delivers time over the Internet—can be provided through peering connections or by developing a substitute.
One service that critically depends on transit is Resource Public Key Infrastructure (RPKI), which has become a crucial technology to secure routing. While some RPKI data can only be acquired through transit, there is no architectural/fundamental reason.
Learn more about the importance of routing security and how the Internet Society is supporting the community to bolster it.
The CDN Perspective
The typical network strategy for CDNs is to expand peering at IXPs and multi-tenant office spaces, as well as penetrate edge/telco networks by deploying on-net caches.
Traditionally, this has made CDNs reliant on transit providers, particularly to help them reach regions not covered by the CDN’s internal backbone. However, this reliance is decreasing. Every dollar not spent on transit can be invested in another part of the service. In some parts of the world customer/hosting companies are not using direct peering and transit is needed to reach them.
The dependency is also different for some infrastructural services, since many big CDNs position themselves as providers of such services, for example, services on a public DNS resolver rely on the global DNS and on the classic Internet. There also seems like a significant reliance on transit for delivering RPKI and, to a lesser extent, NTP services.
Is the Classic Internet Still Relevant?
The short answer is yes. It’s clear that the dominance of the classic Internet—an Internet where connectivity mostly relies on transit over many hops—is decreasing. At what rate and to what effect still needs further research.
We also recognize the research’s strong Western European bias, which warrants further investigation. We used global transit as a measure of the classic Internet, but it may not be a true representation.
Note that we are touching on one of the critical properties of being a part of the Internet, where you need to connect and contract with one party, a transit provider, to be considered a part of the global Internet. Should transit further decline, the Internet likely becomes a less accessible infrastructure for those that do not play the content delivery game.
This post is a summary of our findings which we presented in a keynote paper at the SIGCOMM FIRA workshop and the AMS-IX IP workshop.
Photo by Eugene Zhyvchik on Unsplash