It is 1990: the Internet has a few million users and the first commercial companies have recently adopted this new distributed infrastructure.
The routing of network traffic from one region to another generally depends on the major transit providers (level 1). These levels 1 are at the top of the hierarchy, composed of a few thousand existing AS, forming what is called the network of networks.
A lot has changed since those early days, when small ASs paid the biggest for connectivity. This dependence on intermediaries has resulted in transit costs, indirect routes, long round trip times and a general lack of control over the quality of service. The bypassing of intermediaries by direct peering interconnections became the obvious answer, and Internet Exchange Points (IXPs) then appeared as the default solution for establishing connections.
Between 2008 and 2016, the number of IXPs and members almost tripled. At the same time, accessibility via these facilities has stagnated at around 80% of the announced address space (IPv4) while resilience has increased due to increasing redundancy.
In almost all regions, particularly in Europe and North America, IXP members have grown richer with an increasing number of members and greater accessibility. However, the regional ecosystems were distinct. For example, European IXPs had the largest number of members but the smallest AS (in terms of accessibility), Asia-Pacific was at the opposite extreme.
This growth raises the question of the observable impact of IXPs on the Internet. To answer this question, Queen Mary University in London, in collaboration with researchers from Roma Tre Univ, the GARR Consortium and the University of Tokyo, extracted a large collection of traceroutes covering the same period and identified IXPs crossed.
The IXPs have had a clear impact on reducing the average length of access paths at AS level, particularly for large (hypergiant) global networks. Given that these networks are traffic-intensive, it is likely that a large proportion of Internet traffic has benefited from a substantial reduction in the number of AS crossed.
They have also clearly helped to bypass level 1 transit providers. However, their impact on reducing the number of transit links (not necessarily level 1) visible on the route is more moderate.
Despite these changes, a clear hierarchy remains, with a small number of networks playing a central role. It is interesting to note that there is a small group of very central networks, regardless of whether the paths cross an IXP or not.
In addition, the Internet hierarchy has changed: large central networks have reduced their use of public peerings while IXPs have been adopted by smaller and less central ASs. This is probably due to the increasing popularity of private network interconnections (NIBPs), which are generally favoured by AS when large volumes of traffic are exchanged.
Overall, the increase in the number of IXPs since 2008 has had a clear impact on the evolution of the Internet, shortening paths (mainly) to hypergiants and reducing dependence on Tier 1 transit providers.
The results must be interpreted in the light of the constraints of existing data, and there are a number of areas where work is possible. For example, topological data are independent of traffic volumes and total visibility on the Internet is impossible to achieve.
In addition, content distribution network (CDN) redirection strategies are not included in the traceroutes; it is assumed that accounting for the increasing traffic volumes delivered by these networks would likely support these conclusions.
Source : RIPE