Loading field map
Routing and internetworking is the subfield of computer networks concerned with how data moves across interconnected networks—how a packet originating on one network finds its way to a destination on another, possibly distant, network. It studies the algorithms, protocols, and architectural principles that enable global connectivity, from a small corporate intranet to the worldwide Internet. The central question is: given a collection of networks with different technologies, topologies, and administrative domains, how can data be delivered reliably, efficiently, and fairly from any source to any destination?
The subfield emerged from a practical necessity. Early computer networks were isolated: a local area network (LAN) in one building could not exchange data with a LAN in another, because each used its own addressing scheme, frame format, and medium access method. Internetworking is the solution to this fragmentation. It introduces a common network layer—the Internet Protocol (IP) in the dominant architecture—that sits above the diverse link-layer technologies (Ethernet, Wi-Fi, fiber, satellite) and below the transport layer (TCP, UDP). The network layer provides a uniform addressing scheme and a best-effort delivery service, while routing determines the paths that packets take through this interconnected mesh.
The stakes are high. Routing failures can partition the Internet, making large regions unreachable. Inefficient routing wastes bandwidth and increases latency. Poorly designed routing protocols can cause instability, where routes oscillate or converge slowly after a failure. Because the Internet is composed of thousands of independently operated networks (autonomous systems), routing must also respect policy: an ISP may prefer to route traffic through a partner rather than a competitor, even if the competitor’s path is shorter. Routing and internetworking thus sits at the intersection of algorithm design, protocol engineering, distributed systems, and economic policy.
The core of the subfield is the design of distributed routing algorithms. These algorithms run on every router in a network, exchanging information so that each router can build a forwarding table telling it which neighbor to send a packet to for a given destination. Two major families of algorithms have dominated, each addressing the problem of distributed path computation in a fundamentally different way.
Distance-vector routing is the older tradition. Each router maintains a table of the best known distance to every destination and the next hop to use. Periodically, it sends its entire table to its immediate neighbors. A neighbor receiving the table checks whether any of the advertised paths, combined with the link to the advertising router, offers a shorter path to a destination than its current one; if so, it updates its table. This is the Bellman-Ford algorithm applied in a distributed, asynchronous setting. The approach is simple and requires little memory or computation per router. However, it suffers from the "count-to-infinity" problem: when a link fails, routers may slowly converge on a new path, incrementing the distance metric in a loop until a large threshold is reached. The Routing Information Protocol (RIP), used in small networks, is a classic distance-vector protocol. Its limitations in larger, more dynamic networks motivated the development of the second tradition.
Link-state routing takes a different approach. Each router discovers its directly connected neighbors and the cost of each link, then floods this information to every other router in the network. Every router now has a complete map of the network topology. Using Dijkstra's shortest-path algorithm, each router independently computes the shortest path to every destination. This eliminates the count-to-infinity problem and converges much faster after a topology change. The cost is greater complexity and memory: each router must store the full topology and run a more expensive computation. The Open Shortest Path First (OSPF) protocol, widely used within large enterprise and ISP networks, is the dominant link-state protocol. Intermediate System to Intermediate System (IS-IS) is another, historically used in large service provider backbones.
These two traditions are not merely historical; they coexist. Distance-vector protocols, in their modern form (e.g., the Enhanced Interior Gateway Routing Protocol, EIGRP), incorporate mechanisms to prevent loops and speed convergence, blurring the line. But the fundamental trade-off remains: distance-vector is simpler and scales to very large numbers of destinations if the network is stable, while link-state offers faster convergence and better control over routing loops at the cost of more state and computation.
The algorithms above assume a single administrative domain—a network under common management, where all routers cooperate to find the shortest path. The Internet, however, is a network of networks. Each autonomous system (AS) is independently operated and has its own routing policies. An AS may want to control which traffic it carries, preferring certain paths for economic or security reasons, and it may not want to reveal its internal topology to competitors. This is the domain of interdomain routing, and it is governed by a single protocol: the Border Gateway Protocol (BGP).
BGP is a path-vector protocol, a third tradition that evolved from distance-vector ideas. Instead of exchanging just a distance metric, BGP routers exchange full paths (sequences of AS numbers) to each destination. This allows a router to detect and avoid loops by checking whether its own AS already appears in the path. More importantly, it enables policy-based routing: a router can apply local rules to decide whether to accept a route, which routes to prefer, and whether to advertise a route to neighbors. For example, an ISP might prefer a customer route over a peer route, or it might refuse to carry transit traffic between two of its peers.
BGP is not designed for optimality or fast convergence. It is designed for policy flexibility and scalability. The global routing table contains hundreds of thousands of prefixes, and BGP’s incremental updates, while slow to converge after a major failure, keep the overhead manageable. The protocol’s behavior is deeply influenced by the economic relationships between ASes: customer-provider, peer, and sibling. These relationships shape the flow of traffic and the stability of the global routing system. BGP’s limitations—slow convergence, vulnerability to misconfiguration, and lack of built-in security—are active areas of research and incremental improvement.
The subfield is organized around a clear architectural separation: interior gateway protocols (IGPs) operate within an AS, while exterior gateway protocols (EGPs) operate between ASes. This separation is not arbitrary. IGPs can assume a cooperative environment and optimize for a single metric (e.g., link cost). EGPs must handle policy conflicts, trust boundaries, and the sheer scale of the global Internet. A router typically runs both: an IGP to learn paths within its own AS, and BGP to learn paths to destinations outside. The two interact: a BGP router may use the IGP to determine the next-hop reachability for an external route, and it may inject some external routes into the IGP for internal distribution.
This two-level hierarchy is a durable architectural principle. It allows each AS to choose its own IGP (OSPF, IS-IS, EIGRP, or even static routes) without affecting the global system, as long as it speaks BGP at its borders. The separation also contains failures: a routing loop or misconfiguration inside one AS does not directly propagate to the rest of the Internet.
Routing cannot be understood without its companion: forwarding. Routing is the control plane—the distributed computation that determines paths. Forwarding is the data plane—the per-packet action of looking up the destination address in the forwarding table and sending the packet out the correct interface. Early routers performed both in software, but modern high-speed routers use specialized hardware (TCAM, ASICs) to forward packets at line rate, while the routing protocols run on a separate control processor.
The design of the forwarding table is tightly coupled to the addressing scheme. The Internet uses hierarchical IP addresses: a prefix (e.g., 192.0.2.0/24) identifies a block of addresses, and routers forward based on the longest prefix match. This hierarchy enables aggregation: a router can advertise a single prefix covering many smaller networks, reducing the size of the global routing table. The tension between aggregation and the need for multihoming (connecting to multiple ISPs for redundancy) is a persistent challenge. Multihoming often forces an AS to advertise more specific prefixes, fragmenting the address space and increasing the routing table size.
The subfield today is mature but not static. The core protocols—OSPF, IS-IS, BGP—have been stable for decades, but they are being stretched by new demands. Traffic engineering, once done by tweaking link costs in an IGP, now often uses Multiprotocol Label Switching (MPLS), which sits between the network and link layers. MPLS allows routers to establish explicit label-switched paths, giving operators fine-grained control over traffic flows independent of the underlying IP routing. Segment Routing, a more recent evolution, simplifies MPLS by encoding the path as a list of segments (nodes or links) in the packet header, reducing the state that routers must maintain.
Software-defined networking (SDN) has introduced a different paradigm: separate the control plane from the forwarding plane entirely, centralizing routing decisions in a controller that has a global view of the network. This is a departure from the distributed algorithms that defined the subfield for decades. SDN is most successful in data-center networks, where a single operator controls the entire fabric and can optimize routing for specific traffic patterns (e.g., incast, elephant flows). In the wide area, where administrative boundaries and policy complexity dominate, distributed routing remains essential, though SDN principles are being applied to traffic engineering within a single AS.
Security has become a central concern. BGP was designed without authentication, making it vulnerable to route hijacking (an AS falsely announcing a prefix) and route leaks (an AS propagating a route it should not). Efforts like Resource Public Key Infrastructure (RPKI) add cryptographic validation of prefix ownership, but deployment is gradual. The subfield now includes a significant body of work on securing the routing infrastructure, detecting anomalies, and designing protocols that are robust to misconfiguration and attack.
The fundamental questions remain: how to balance optimality with scalability, policy with performance, and distributed autonomy with global stability. The answers are embodied in the protocols that run the Internet, and the subfield continues to evolve as the network grows in scale, diversity, and economic importance.