TWO
A Taxonomy of "Fragmentation"

On one side we have advocates of "data sovereignty;" on the other we have the critics of Internet "fragmentation" or "Balkanization." One can bring clarity to this debate by asking a simple question: what does fragmentation of the Internet mean? What are the conditions that have to prevail for us to call the Internet "fragmented" or "not fragmented," "Balkanized" or not? The literature on this topic contains very few attempts to grapple with that question, and most of them are weak or muddled conceptually.

Let's start with basic definitions. As a noun, the word "fragment" implies a part that has been disconnected from a greater whole, like a shard of glass detached from a broken bottle. As an adjective, "fragmented" means existing or functioning as though broken into separate parts, disunified, dismembered.

How would one recognize a fragmented Internet when one sees it? Herein lies the central irony of the Balkanization debate. One could use the concept of fragmentation to arrive simultaneously at two diametrically opposed conclusions:

• The internet is now and always has been fragmented
• The Internet is not now and never will be fragmented

The unifragged Internet

The Internet could be characterized as fragmented because it was designed to be a network of networks. The basic units of internetworking are known as Autonomous Systems. Autonomy is a strong word; it implies an ability or freedom to act independently. In a political context, it refers to the ability to govern oneself. In a network context, it means something similar: the ability to set policies for naming, addressing and routing, and to control or manage many other aspects of network operations. In this respect, the Internet is already "Balkanized." It is a federation of Autonomous Systems with an extensive capability for selective, fine-grained "secession" from practically any other part of the federation.

The Internet is unlike the Balkans in one crucial respect, however: all Autonomous Systems speak a common language. That language is a set of data formatting, naming, addressing, and routing standards collectively known as "the Internet protocols," the most basic of which is Internet Protocol (IP). Even so, nothing compels any single Autonomous System (APs) to open itself up completely to all others. All APs can exercise - or try to exercise - control over who they interconnect with, what packets they admit into or out of their systems, what services they want to accept or block, what content can enter and leave. And it is clear that many if not most Autonomous Systems are exercising this capability. They are blocking domains associated with attackers, filtering IP address ranges associated with spammers, using firewalls and security software to detect and block malware. In this sense, every Autonomous System is a "fragment" of the whole Internet, a part distinct from a greater whole. As one scholarly duo put it, the Internet is always "locally configured as well as globally networked" (Lobato and Meese, 2016, p. 14).

"The Internet is already 'Balkanized' - its networks are capable of selective, fine-grained 'secession' from practically any other part of the system."

At the same time, we could say that the Internet is not now and is unlikely ever to be truly fragmented, if "fragmentation" is understood in the harder sense of "broken into separate parts" with insurmountable barriers to communication between them. As noted before, all Autonomous Systems on the Internet use the same protocol for data communication: IP, which includes a globally consistent address space. IP functions as the lingua franca that connects and makes compatible an enormous number of lower-level physical layer technologies (such as copper, radio, or fiber) and higher-level transport and application standards (such as the Web protocol, word processing applications, or instant messaging apps). This is often visualized as an hourglass, with IP being the single, common protocol that converges a growing and constantly changing array of different technologies above it and below it in the protocol stack.

People use IP because everyone else is using it. The benefits of common usage of it are so great, the utility of having at least the potential to exchange information with anyone in the world at any time is so enormous, that it is difficult to imagine any organization refusing to use IP for data networking.

Far from splintering or faltering, Internet connectivity is spreading virally. While every Autonomous System has the ability and the right to manage the nature of their Internet access, openness to the rest of the world's computers is the default value in this environment. If Autonomous Systems want their network to limit or restrict access, they have to do a lot of work to make it happen.

Figure 1: Hourglass Model for Internet Protocol

The Internet is at once unified and fragmented. It is unifragged.

A taxonomy of "fragmentation'

To appreciate the policy implications of this paradoxical mixture of connection and disconnection, access and blockage, fragments and the whole, we need a finer vocabulary. Consider the following scenarios:

• A cable is accidentally cut and suddenly thousands of people are left without Internet access for several hours; is this what we mean by fragmentation?
• A friend provides one with a link to what he calls a hilarious YouTube video. But when you click on it you see only a blank screen with the message, "This video is not available in your country." Is this fragmentation?
• The government of a country orders its ISPs to block domains or URLs associated with content or services deemed subversive or illegal; is this fragmentation?
• A major online newspaper blocks access to its content unless you pay them a monthly subscription fee; is this fragmentation?
• Two major ISPs stop peering with each other and for a day or two, packets from some of their customers don't know how to find their way to their intended destination; is this what we mean by fragmentation?

In some sense, all of these result in something that might be called fragmentation in the sense of interrupted communication paths. But blockages and downtimes are constantly coming into and going out of existence on the Internet (and on any network). Some of these are routine, inevitable, and even beneficial aspects of a distributed system. Others are unintended, accidental, objectionable, and pathological. Clearly, there are crucial qualitative differences among them. Figure 2 provides a diagram that helps to sort them out.

The most basic difference is whether the blocking or disruption is intentional or unintentional. Intentional blocking means that someone has made a conscious decision to limit or manage their exposure to the global Internet. It involves things like spam filters, firewalls, or lists of blocked URLs. Unintentional fragmentation arises from breakdowns or technical incompatibilities or physical barriers that make different parts of the Internet literally incapable of interoperating. The cut cable, the temporary loss of connectivity caused by malfunctioning equipment or downtime; the use of protocols or applications that simply don't work together - all are examples of technical incompatibilities that are (usually) unintended.

Unintentional fragmentation means: You can't get there from here. Intentional fragmentation means you can get there from here, but someone has decided not to let you.

Insisting on the distinction between intentional and unintentional action focuses attention on four critical questions for Internet governance. If the blocking is intentional, who made that decision, whose interests does it serve, who is affected by it, and how accountable is the decision maker to the persons affected?

Figure 2: Typology of "Fragmentation"

Figure 2 shows three basic categories of intentional forms of network management that might be construed as fragmentation: (1) private, voluntary decisions made by organizations regarding their own facilities and services; (2) decisions made by private intermediaries offering service to the public, which affect not only their own property but the capabilities and exposure of their customers; and (3) decisions imposed on private networks, public intermediaries, and users by third parties such as governments; these barriers or filters arise from policy decisions and are coercive in nature.

A key distinction between intentional and unintentional fragmentation has to do with its durability. Unintentional barriers to communication tend to be temporary, and thus a less significant part of the Internet's structure. If the limitations are a product of malfunctions or breakages, they are perceived as problems to be corrected, and are quickly fixed most of the time. If they are a product of what I call structural technical incompatibilities, they last longer but are perceived as bothersome and tend to...