Bärwolff's 2010 PhD Thesis "End-to-End Arguments in the Internet: Principles, Practices, and Theory"
- Bärwolff (2010) End-to-End Arguments in the Internet: Principles, Practices, and Theory (4.2M)
- · Available as a printed book on Amazon's CreateSpace as well as on Amazon.com ($25, scaled by .85, fairly narrow margins – but, in all, a really pleasant read); as of May 2011 the book is also on international Amazon sites. Note that if you want to do me a favor (as opposed to Amazon), order the book with CreateSpace.com, the profit margin for me is largest there. Finally, see the continuously updated errata, mostly for the CreateSpace book.
- · Also as a super-long-term officially archived 15MB version here (featuring a print-on-demand option, for some €40)
- · For prior versions, changelog, plus sha256 hashes see this folder listing
The end-to-end arguments — a founding principle of the Internet — have inspired a world of opinionated interpretations, rearticulations, and comments about their lower-level technical and higher-level normative merit. However, their precise meaning and scope of applicability have remained diffuse — arguably as diffuse as the constituency of stakeholders in the very Internet itself. Our thesis elaborates the end-to-end arguments by putting them into a meaningful context of other principles as well as the current realities of the Internet. Also, it elaborates their normative content in view of a defendable set of purposes. To these ends we have conducted a most thorough study of primary literature going back to the intellectual predecessors of the Internet in the early 1960s, and developed the several notions of this thesis in a largely desk-based research effort.
Following a comprehensive discussion of the various different versions of the end-to-end arguments both prior to and subsequent to the seminal Saltzer, Reed, and Clark (1981) formulation we trace the actual manifestations of the end-to-end arguments walking from the Arpanet as developed in the late 1960s to the eventual Internet architecture emerging from the mid-1970s. We find that the descriptive content of the end-to-end arguments is, in fact, broader than their self-contained formulation as a logical argument about application completeness implies. Second, we find that they are best conceived as one principle within a framework of other, no less relevant principles. And, third, we argue for a revised normative take on the end-to-end arguments that emphasizes the importance of edge redundancy as a crucial means of extending their scope beyond the classic considerations of data integrity alone. In all, we add much needed focus and clarity to a notion that has been carried so far away from its original content as to cloud its true relevance in both today’s Internet and tomorrow’s.
Beyond offering a comprehensive descriptive account of the very notion of end-to-end arguments (Part I) and their seminal manifestations in the history of the Arpanet and Internet (Part II), a number of useful abstract themes have emerged (particularly from Part III):
The end-to-end arguments are more than a self-contained blackboard exercise about the logical correctness and completeness of data transfers. It is especially in conjunction with other principles that the end-to-end arguments extend well beyond cases that obviously collapse into the application end points and require from other entities no level of service greater than any epsilon greater than zero. We have here argued that functions like fragmentation and congestion control have moved to the end points not because of completeness and correctness concerns, but because of the infeasibility of maintaining a meaningful explicit interface for such functions across trust boundaries (section 7.1). Moving from the descriptive to the normative, we have also argued that functions such as throughput maximization and delay minimization can be moved more firmly to the end points if we pursue the normative notion of ‘edge redundancy’, allowing users and Internet attachment points to multiplex on each other (section 8.2).
There is substantial merit in distinguishing between vertical end-to-end arguments and horizontal ones, only the latter of which speaks to the application level structure of today’s Internet. The crucial difference between the two is that while the vertical arguments logically
entail the notion of irreducibly minimum common ground at the shared inter-network level so as to reduce the adverse effects from negative externalities, the horizontal ones allow functions to float freely between end users and network intermediaries subject to trust and other pertinent trade-offs, for the effects of such sophistication are limited only to the entities part of any such structure. The minimality of the IP protocol is an impressive testament to the power of the vertical arguments. The horizontal arguments, on the other hand, have forever been far more malleable to application specific concerns at the very application (protocol) level. While this notion is not original to our thesis, we have offered an in-depth discussion and elaboration, particularly in section 3.3.
The end-to-end arguments have been notoriously prone to be interpreted pursuant to a vague set of higher-order purposes (innovation, economic growth, and democracy) which together with the widespread confusion between vertical and horizontal arguments has led to a number of nonsensical (policy) implications. With a nod to Hayek (1973), we argue that it is crucial to appreciate that only the purposes overall order and individual liberty can be solidly defended and used as a measure to judge any version of the end-to-end arguments (or, indeed, any other principle or rule). On a very much related note, the appreciation of tussles in the Internet, along with the understanding that they can only be contained but not eliminated, does not go together well with any strict and dogmatic version of the end-to-end arguments. What it does go together well with, however, is the edge redundancy version of the end-to-end principle lending further credence to the notion of tussle separation — a notion we have developed particularly in chapter 8.