What network architecture means
A network architecture is a plan for how the parts of a network fit together: what jobs are done where, which rules the parts follow, and how new parts can be added. It matters because networking is a large problem. A single program that handled cables, error correction, routing, reliable delivery and applications all at once would be hard to build, test or change. Network designers handled this by dividing the work into layers.
In a layered design, each layer does one job and relies on the layer below it to do its own. The lowest layer deals with signals on a physical link. The next one handles moving data reliably across one link, another handles finding a path across many links, another ensures that data arrive complete and in order, and the top layers deal with applications such as mail or file transfer. A layer can be replaced without rewriting the others, much as one can change a road surface without redesigning the cars that drive on it.
The era of closed systems
In the early 1970s, the major computer makers began offering networks for their own products. IBM introduced Systems Network Architecture (SNA) in the mid-1970s, built mainly around large central machines and many terminals. Digital Equipment Corporation developed DECnet for its minicomputers, and other companies had their own designs. Within each family, the products worked together well. Between families, they mostly did not.
For customers, this created lock-in. A company that had bought from several makers might end up with islands of equipment that could not exchange data, or only through awkward converters. Governments and large buyers pressed for common standards, partly to widen their choices.
Two layered models
Two important layered models emerged from the 1970s and early 1980s.
The OSI model. The International Organization for Standardization began work on Open Systems Interconnection in the late 1970s. Its reference model divided networking into seven layers: physical, data link, network, transport, session, presentation and application. The model was published in the first half of the 1980s. Many governments supported OSI, and networking textbooks adopted its seven-layer vocabulary, which people still use when they say a problem is at “layer two” or “layer three.”
The TCP/IP suite. A separate effort began in the United States among researchers working on the ARPANET and on connecting different kinds of networks. Its protocols, TCP and IP, divide the job into fewer layers, often described as link, internet, transport and application. The design focused on a practical goal: letting any network join the larger internet by implementing the common protocols, regardless of the machines attached. The story of those protocols is explored in the history of the internet protocols.
The standards contest
During the 1980s, the two approaches were seen as rivals, and commentators sometimes called the dispute the protocol wars. OSI had formal backing from standards bodies and governments, and its supporters argued it was more complete and carefully designed. TCP/IP had working software, widespread use in universities, and implementations that were distributed freely or cheaply, particularly in Unix systems. Because people could use it right away, adoption grew.
Historians and engineers describe the outcome in different ways, and reasons given for TCP/IP’s success include its early working code, lower complexity, and good timing as the academic and commercial internet expanded. OSI’s drawbacks, as often cited, were that standards took a long time to complete, were complicated, and had fewer products. It would be an overstatement to say OSI failed entirely, since its seven-layer model remained an influential teaching tool and some OSI-derived pieces survive in other technologies. By the early 1990s, TCP/IP had become the common choice for connecting different networks.
Interconnection in practice
How did separate networks come to interconnect? Several techniques were used.
- Gateways. A gateway is a device or program that translates between two networks that use different rules. In the 1980s, mail gateways let users on different systems exchange messages, though addresses were often awkward.
- Encapsulation. One network’s packets can be carried inside the packets of another, so that a standard protocol can ride on top of a proprietary or older link.
- Common addresses and names. The internet’s address scheme and the domain name system gave a shared way to identify machines.
- Vendor support. Eventually computer makers added TCP/IP to their own products, often alongside their older networks. IBM, DEC and others did so during the 1980s and 1990s.
This means the single global internet was not created by throwing out all earlier networks. Many were gradually connected, converted or retired.
A design idea: the end-to-end principle
One influential architectural idea associated with the early internet is that the network itself should do relatively little, simply moving packets, while the computers at the ends handle tasks such as reliability. The idea, discussed by researchers in the early 1980s as the end-to-end argument, made it easier to add new services without changing the middle of the network. It also meant that the core network had few built-in safeguards, which later turned out to have consequences for spam and for computer security.
Why it matters
Layering and open protocols can seem like technical details. They explain, however, why a smartphone, a laptop and a game console from different companies can all reach the same site, and why a new application can appear without waiting for the cable network to approve it. The same structure underlies local area networks, the World Wide Web and the Internet of Things. When people say that the internet was built on open standards, this history of architecture is what they mean.