Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more

TCP/IP connected unlike networks by keeping the network layer’s job small: Internet Protocol (IP) moves addressed datagrams between networks, while end systems add the reliability and ordering that many applications need. That division let networks with different technologies and operating characteristics interoperate without first adopting one shared internal design.

Why connecting unlike networks called for a pragmatic design

The central challenge was not to build one enormous uniform network. It was to connect networks that had been built separately and did not share a physical technology or a common reliability mechanism. Requiring every network to work identically would have made interconnection depend on changing all the networks underneath it.

In spring 1973, Bob Kahn and Vint Cerf began detailed protocol design as part of the broader internetting effort. Their first written version circulated as INWG #39 in September 1973; a more developed account appeared in 1974. The history is collaborative, rooted in work across the International Network Working Group, rather than a lone-inventor story. The Internet Society’s history of the Internet describes that development and the architectural approach.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

IP’s small job is what makes interconnection possible

IP provides addressing and datagram delivery across interconnected networks. A datagram is handled independently; IP does not promise that it will arrive, arrive only once, or arrive in order. Nor does IP itself provide end-to-end flow control. RFC 791 states the boundary plainly: “The internet protocol does not provide a reliable communication facility.” RFC 791, published in September 1981, is the specification for that design.

This is not a missing feature accidentally left unfinished. A minimal common service makes it possible to carry traffic across networks whose local protocols, packet-size limits, and performance differ. IP gives them a shared way to address and forward datagrams without insisting that every network use the same machinery internally.

TCP supplies reliability at the endpoints

When an application needs a reliable, ordered stream of bytes, Transmission Control Protocol (TCP) can provide it above IP. TCP numbers data, uses acknowledgments to track what has arrived, and retransmits data when acknowledgments do not arrive. Receivers can reorder segments and discard duplicates; flow control helps a receiver communicate how much additional data it can accept.

The important placement choice is that this recovery work happens at the hosts, not as a universal promise made by every network along the way. The original TCP specification, RFC 793 (September 1981), describes TCP over a potentially unreliable datagram service and qualifies reliability on the endpoints continuing to function and the internetwork not becoming completely partitioned. RFC 793 is historical: the RFC Editor identifies it as obsoleted by RFC 9293, so it is useful here to explain the architecture’s origins, not as current implementation guidance.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

How the architecture copes with variation and failure

Networks differ in bandwidth, delay, packet loss, reordering, and maximum packet size. TCP/IP does not make those differences disappear; it allows the internetwork to function without requiring them to be identical. RFC 1122’s host requirements describe gateways forwarding IP datagrams independently, while reliability and flow-control state lives at endpoints. RFC 1122 (October 1989) presents this as an architectural goal.

Independent forwarding also makes it possible for alternate paths to help maintain service when an individual gateway or network fails. This is resilience through a network of networks, not a guarantee that every failure will be invisible: a complete partition or failed endpoint can still prevent communication.

Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

Pragmatism, not perfection, explains TCP/IP’s durability

TCP/IP’s achievement is a useful division of responsibility: the internetwork layer offers a common, modest delivery service, and endpoints implement the communication behavior their applications need. That makes the architecture adaptable, but it does not mean the modern Internet is purely end-to-end or that the original principles solve every later problem.

The Internet Architecture Board’s RFC 1958 (June 1996) calls the Internet’s growth evolutionary rather than the result of a grand plan, and recognizes that architectural principles change as technology does. Its discussion reflects pressures including scaling, speed, multimedia, quality of service, and security. The enduring lesson is not that TCP/IP was perfect from the outset; it is that a deliberately limited common foundation could grow and adapt.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.