Capítulo 5. Conclusiones y recomendaciones
A. Anexo: Formato utilizado en la prueba piloto
The Internet can be defined as a dynamic and self-organised network of inter-connected networks. The ecosystem of the Internet is said to involve agents with a diverse set of functional roles and objectives. Hence, the Internet is composed of thousands of Internet
Service Providers that are operating different parts of this Information and
Communication Technology (ICT) infrastructure, providing services to access and use the Internet. The services of ISPs are typically used by content providers and end-users (including machines or ‘bots’, which are using the largest part of the Internet infrastructure). The agents composing the Internet ecosystem include:
• Access Providers: Internet Service Providers selling Internet Access to individuals and / or business customers (e.g. mobile broadband operators such as Aircel or Vodafone).
• Transit Providers: considered as geographically distributed large backbone network operators, which were historically, and are still presently, paid to transfer traffic over large distances. A transit provider might also be an access provider (e.g. Level 3 Communications or Cogent Communications).
• Content Providers: Internet Service Providers that generate the content for end- users on the Internet. Content Providers include providers of information, video, e-Commerce, social networking or search results, amongst others (e.g. Google, Facebook or Netflix).
• Content Distribution Networks (CDN): these are Internet Service Providers that store customer content locally for a quicker fulfilment of download requests from nearby users (e.g. Akamai). Their customers are usually access providers. • Internet Exchange Points (IXP): IXPs operate facilities of (paid) inter-
connection, where other Internet Service Providers may be present and inter- connect with other ISPs (e.g. the London Internet Exchange (LINX), or the Amsterdam Internet Exchange (AIX)).
Internet Service Providers engage in a set of formal and / or informal relationships with
each other, providing access to value added end-user services, by collectively routing Internet traffic. Such relationships (and especially their routing policies, see section 2.3) determine constraints to the respective paths through which Internet traffic might flow and therefore have implications on the robustness and further engineering of the Internet. Insight 1: Given the coexistence of formal and informal relationships amongst Internet
Service Providers, we expect their clear identification to be quite challenging. While
routing policies are often available (see above), ISPs are likely to maintain confidentiality about their business interconnection practices. Hence, we argue that available (secondary) data still contains many unidentified Internet Service Provider relationships, leading to the present difficulties, for current research, in providing a satisfactory picture of the Internet infrastructure.
The carriage of traffic on the Internet is usually organised through packet switching networks. This method relates to the transfer of data, which is split into smaller data packets for simultaneous transaction purposes. The carriage of these data packets through a network follows approximately symmetric transactions. This means that the senders and receivers of a given data packet are involved in the same amount of data transactions. Hence, these transactions may involve symmetrical payments, where both the sender and the receiver pay transaction costs to their Internet Service Providers. As described in detail below, an ISP may then pay their partners for the transit of the data packet in the upstream Internet, assuming they do not have a peering relationship with them (Woodcock, 2003). Following this organisation, the data packets are routed through vertically related Autonomous Systems, belonging to one or more Internet Service
Provider(s) along the path to the final data packet destination. Internet routing
mechanisms facilitate this exchange of data packets through computer networks using the so-called TCP/IP stack, which was jointly developed by Cerf and Kahn (1974) through DARPA funding, representing a cornerstone of the information-based Internet. In detail, the Internet Protocol establishes interconnections with the aim of delivering given data packets from sender to receiver, while both these agents obtain a unique Internet Protocol (IP) address. Such IP addresses are organised in address ranges (prefix) managed by Autonomous Systems. Hawkinson and Bates (1996, p.2) suitably define an Autonomous System as
‘… a connected group of one or more IP prefixes run by one or more network operators which have a single and clearly defined routing policy’.
Each Autonomous System is therefore a physical collection of bare-metal (gateway) routers, represented by a unique IP address prefix (see Glossary). These IP address prefixes are under one common administrative control by the Autonomous System. However, multiple Internet Service Providers may share the ownership, and hence the administrative control, over an AS. Such policy-based routing techniques between IP address prefixes represent interconnections that are established between a pair of Autonomous Systems in packet switching networks, the ‘lifeline of telecommunication services’ (TRAI, 2016f). The International Telecommunications Union refers to peering or transit as relationships between Autonomous Systems (ITU, 2007, pp.7-9). Peering relates to an exchange of traffic between a defined set of Internet Protocol networks usually at no charge, except for paid peering (Norton, 2011). This exchange of traffic takes place mostly when Autonomous Systems share the same traffic volume characteristics. In any peering relationship, both sides agree to the peering conditions, which might include network coverage, operations, and maintenance of the network, as well as the volume of traffic that can be exchanged. While the process for engaging in peering relationships is often undisclosed, some Internet Service Providers share the peering policies for their Autonomous Systems more openly, see e.g. the peering policy of the Swiss National Research and Education Network (SWITCH, 2016). Some efforts are undertaken to collect the routing policies to the Internet through the Internet Routing Registry (IRR), a distributed routing information database formed in 1995 (IRR, 2016). The distribution and organisation of IP addresses is managed at continental level through the Réseaux IP Européens Network Coordination Center, the regional Internet registry for Europe, which also collects routing information using RCC, the Remote Route Collector (RIPE NCC, 2016a). The Asia Pacific Network Information Center (APNIC), the Regional Internet Registry of Asia, also engages in the collection of routing information (including Indian ones) through the above-mentioned Internet Routing Registry (APNIC, 2016). Nevertheless, when the ASes of two Internet Service Providers enter into a relationship, the type of contract arrangements depends largely on the balance of contributions that benefit both parties (ITU, 2007). If a peering arrangement is not possible, the Internet Service Provider parties might engage in transit arrangements, which allow them to reach all remaining parts in the Internet periphery. Here, larger
other ISPs. Once a transit arrangement is set, the sender pays the full costs of interconnection. The charges for these interconnections are usually undisclosed and negotiated on commercial terms. As previously shown, transit arrangements with one of the large Tier-1 Internet Service Providers, or those that directly connect to the Internet backbone, can provide a smaller Internet Service Provider with access to the rest of the Internet, while also introducing costs leading to potentially high global connectivity prices (ITU, 2007, p.9; CAIDA, 2016a). By gaining such access to the rest of the Internet, a smaller Internet Service Provider would be reliant on those larger Internet Service
Providers for the purpose of global internetworking to and from the Internet periphery,
interconnecting end-users. The small number of large upstream Internet Service
Providers, with a strong interconnection demand from downstream ISP,s forms the higher
hierarchical structure of the Internet, the core, holding great negotiating power over interconnection practices and prices with the smaller downstream ones (D’Ignazio and Giovannetti, 2006, pp.2-13). Peering is common amongst members of the Internet core who also provide global connectivity through paid transit to the other ISPs, forming the lower layers (Tiers) of the Internet (Woodcock, 2003). Furthermore, the oligopolistic structure of the Internet core guarantees the largest Internet Service Providers with unidirectional revenue flows arising from transit payments from smaller Internet Service
Providers residing in the Internet periphery. This asymmetric flow of resources reinforces
the incentives to minimize transit costs for a growing number of Internet Service
Providers and end-users in the Internet periphery and maximises their set of peering
relations.
A large body of literature in Computer Science focuses on the study of the relationships between Autonomous Systems to explain peering and transit relationships. The relationships amongst ASes are considered to have a significant impact on the flow of traffic through the Internet (Subramanian et al., 2001). The Autonomous System roles in these relationships are, according to Alaettinoglu (1996) and Huston (1999), either of
provider-to-customer, customer-to-provider or peer-to-peer nature. This definition is also
used in the work of Gao (2001, p.734), who further states that two Autonomous Systems, which are operated by one Internet Service Provider, may have sibling relationships, where each AS provides transit services for the other. This is especially relevant when considering that one Internet Service Provider might operate multiple Autonomous Systems. Moreover, Gao and Rexford (2000) show that a pair of ASes may peer indirectly through a transit Autonomous System. Gao, Griffin and Rexford (2001) expand on this,
suggesting that a pair of Autonomous Systems may also have backup relationships to provide connectivity in the event of failures or downtimes.
Nevertheless, the structuring of the Internet allows larger Tier-1 providers to obtain central positions on the Internet, given their transit relationships that are crucial for interconnecting traffic, across geographical distances, with / for other Internet Service
Providers. According to Economides (1995, p.678), in economics, structural bottlenecks
in the interconnections occur when an economic agent has a monopoly, and market power, over a link (or relationship) with other economic agents, creating essential facilities within a network. Structural bottlenecks, hence, cause traffic flow congestions in the digital supply chain, which occur when an Autonomous System receives more data traffic than it can cope with. This definition provides a cornerstone for our work. When referring to ‘structural bottlenecks’, we speak of the economical, rather than the technological, definition. This view of Economides (1995) is also identified by Subramanian et al. (2001), stating that relationships between Autonomous Systems have a significant impact on the flow of traffic through the Internet, while hierarchy symbolises business relationships between Autonomous Systems. Subramanian et al. (2001) also state that customers should be at a lower hierarchical layer compared to their providers, a concept which is best captured by using methods of directed network graphs where edge directions indicate the types of relationships between two Autonomous Systems. To better capture the types of relationships between ASes, Luckie et al. (2013) propose the usage of the concept of an Autonomous System’s Customer Cone based on the PhD work of Giotsas (2014). The following Figure 2-1 visualises this concept of a Customer Cone.
Key
Figure 2-1: Internet Service Provider (ISP) Customer Cones.
Considering Figure 2-1 above, the customer cone of ‘ISP4’ would be ‘ISP1’, ‘ISP2’, whereas the customer for ‘ISP5’ would be ‘ISP3’. This is especially relevant since two Autonomous Systems might have a peer-to-peer relationship in one location of the Internet and a provider-to-customer, or customer-to-provider one at another location on the Internet. Nevertheless, and despite this complexity, the business relationships amongst pairs of ASes is a theme where too little economic research has been undertaken to date. However, we consider these relationships as hugely important when studying the formation and structure of the upstream Internet market of mobile broadband operators, since Internet connectivity is not established on the basis of the shortest paths to the final destination but on their economic value.
The fact that an Internet Service Provider may inhabit one or a multitude of Autonomous Systems, whereas an AS may also be shared amongst a set of Internet Service Providers,
adds complexity to the interconnection relationships. This is further complicated by the fact that there is no central authority managing the total number of connecting Autonomous Systems. The organisation of these Autonomous Systems is only vaguely defined. Internet Service Providers that manage ASes usually also provide global connectivity to their customer networks, but this type of connectivity comes in a variety of sizes and structures (MIT, 2009). The shared definition amongst Computer Science practitioners classifies ISPs into three different Tiers, as Figure 2-2 below illustrates. Here, Tier-3 Internet Service Providers are believed to provide connectivity to a low number of geographically local end customers while being connected to upstream Tier-2
ISPs. These Tier 2 Internet Service Providers cover the regional connectivity scope (state,
or region wide), while linking to Tier-1 ISPs for international connections. Hence, Tier- 1 Internet Service Providers capture a global connectivity scope, being able to reach any Autonomous System on the Internet (MIT, 2009). Nevertheless, there are only a handful of these large International Tier-1 Internet Service Providers, stated e.g. in the CAIDA (2016a) AS-Rank. However, the specific literature finds no consensus about the amount of such large Tier-1 ISPs. Importantly, Tier-1 Internet Service Providers are not reliant on buying connectivity transit services from other ISPs but mostly rely on settlement-free peering relationships with other large Internet Service Providers, reciprocally exchanging traffic between each other. This provides them with bargaining power and global connectivity criticality on the Internet. Moreover, the transit services of large Tier-1 ISPs would usually cover priced services that allow smaller ISPs (from Tier-2 or Tier-3) to access the entire Internet through routing agreements. These routing agreements may not be transitive since Internet Services Providers are not obliged to carry traffic to other
ISPs. Moreover, Tier-2 Internet Services Providers are usually peering with some other ISPs but are still reliant on purchasing Internet Protocol (IP) transit (see Glossary) from
Tier-1, or other regional Tier-2 ISPs, depending on the final data packet destination. An
ISP that purchases transit would then be a customer in a customer-to-provider
relationship, as described in their routing policy above. Such a Tier-2 ISP would most likely try to save IP transit costs by establishing peering relationships with as many Tier- 2 or Tier-1 Internet Service Providers as possible. Lastly, Tier 3 ISPs do not usually sell any transit to other ISPs but are entirely reliant on purchasing transit from other Internet
Service Providers in order to reach the entire Internet. The following Figure 2-2 provides
an overview of possible interactions between ISPs and an Internet Exchange Point (IXP) located within the three Tiers.
Key
Figure 2-2: Internet Protocol (IP) Tier networks.
Insight 2: Based on the economic nature of Internet Service Provider relationships, we expect to observe a hierarchical network structure where a low number of globally acting Tier-1 Internet Service Providers provide global connectivity to a larger number of regional Tier-2 and local Tier-3 Internet Service Providers, except for when a given data traffic remains local. While this economic nature is largely agreed upon in the literature (e.g. Luckie et al., 2013), little research seems to focus on the implication of such hierarchical structuring, characterised by strong bargaining powers of a few Tier-1