There are two important prerequisites that any network needs to be successful: protocols and standards. The former refers to the rules that computers — which are the component parts of a network — need to follow to exchange signals and data, and the latter to the implementation of the same procedures across the same network.
Although the first implementers of networks resorted to building their own protocols, there was an understanding that ubiquitous networking could only become a reality if there was a comprehensive set of open standards to support it. To that aim, in 1977 the International Standards Organization (ISO) launched a program to develop general standards, which culminated in the open systems interconnection model, or OSI; from Davies and Bressan:
The technical merits and organization of the OSI standards process is a separate story in its own right, but the influence on the planning of academic networking was profound. Policy makers embraced the concept of open standards as an essential component of open markets; researchers welcomed the promise of vendor independence and open interchange of information; funding bodies welcomed the opportunity for efficient resource sharing. Open systems came to be seen politically as one of the essential elements for providing integration of the European infrastructure.
With this level of open reception initially presented towards the development of OSI standards, it’s worth considering why and how things have changed. More recently, the OSI model was deemed obsolete by the 451 Alliance, a global think tank composed of IT professionals, business leaders, and academics. In a blog post from 2019, the group claims that the model has run its course, its layers are collapsing, and business leaders in the digital era are aligning themselves instead with TCP/IP — another set of protocols that govern the connection of computers to the Internet.
Let’s take each model in turn. The OSI model has been replicated below.
Each layer of the OSI model has a specific network function, dividing the broader task of network interconnection into seven smaller, more manageable tasks. Moreover, each layer is self-contained, and communicates only with the layers above and/or below itself. The transfer of information follows a two-step process: data must travel down the layers of OSI on a sending device and up the layers on the receiving device. A full understanding of each layer (alongside an example of data transfer flows in practice) can be found here.
Alternative proposals to improve upon OSI have been proposed from a cloud computing angle, and Cisco has put forth its own version in which it introduces a new layer called “Services” to denote the foundation of application architectures. Given the value of customized frameworks to reflect different organizations’ capabilities, I expect there will continue to be new models proposed by networking- and Internet-based companies for the foreseeable future.
The next phase of networking
The TCP/IP model divides the process into 4 layers instead. Although its core tasks correspond to existing layers in the OSI model, it provides a consolidated view which some argue more accurately reflects how the Internet functions today. TCP/IP also removes the session and presentation layers from the core framework; while some deem these unnecessary, others have argued that they are incorporated instead in the application and transport layers.
The core argument of the 451 Alliance is that TCP/IP is more flexible and adaptable than OSI when applied to the current digital era. They claim OSI’s focus on a full stack of platforms is outdated since IT is now a solution integration business.
Gone are the days when IT business leaders buy a full stack of IT products and services from one or just a few companies.
In fact, what collapsed the OSI model is abstraction and the emergence of clickable infrastructure. We are in an API, service mesh world, where IT’s main job is to stitch together multiple software building blocks to create a solution. These building blocks are cloud providers, startups, open source code, homegrown code, legacy providers, etc. The glue that ties it all together is of high value.
This point only stands if one considers both models to have the same purpose. OSI was launched as a conceptual framework to describe the functions of a networking system — and as it is rooted in the early stages of network computing, it still maintains broad theoretical applicability. Many conceive of TCP/IP in a more practical sense, used by business leaders and professionals to stitch together the building blocks described above (i.e. cloud providers, open source code…).
While the model has recently gained ground, TCP/IP has its roots in work sponsored by DARPA (an agency whose parent is the Department of Defense) in the 1960s. In a paper on “The DoD Internet Architecture Model,” which would later become TCP/IP, Vint Cerf and Edward Cain contend that the model suffers from architectural incompleteness, fails to explicitly indicate where and how front-end systems can be incorporated, and may not incorporate provisions for concepts in end-to-end and multilevel security. However, the report ends on a proactive note.
It is opinion of the authors that the DoD Internet Model is the most fully developed, military-oriented networking architecture in existence. It is based on over 10 years of field experience with the most advanced packet switching systems in the world. We do not believe, however, that the model can remain static. The many loose ends are proof that the model and its protocols must evolve. It is our hope that this evolution can be accomplished in cooperation with our NATO allies and generally within the framework of the national and international protocol standardization initiatives now underway.
In spite of the different objectives of these two models, there are some important reasons for the rise of TCP/IP. Whereas the layers in OSI are highly dependent upon each other, TCP/IP has a far more flexible architecture, broadening its popular appeal with IT business leaders in particular. But in the end, the main objective of TCP/IP is to help IT specialists determine how specific computers should be connected to the Internet, and best practices for transmission. The OSI model, while providing a valuable foundation for networking theory, was never intended to guide solution integration.