Summary
Highlights
The network layer is crucial for the Internet, implemented in billions of devices and responsible for gluing the Internet together. It's divided into two main parts: the data plane (local router actions like forwarding) and the control plane (network-wide packet coordination).
The data plane section will cover the big picture of the network layer, the internal workings of an Internet router, the IP Datagram header, IP addressing, Network Address Translation (NAT), IPv6, generalized forwarding, and middle-boxes. It will explore both principles and practices.
At the sending host, the network layer encapsulates transport layer segments into IP Datagrams, including addressing information, and passes them to the link layer. At the receiving host, it extracts the payload and demultiplexes the segment to the appropriate transport protocol.
Routers are the primary network layer devices in the core. Their job is to receive and forward datagrams to the correct outgoing link. This raises questions about how routers determine the appropriate outgoing link (local issue) and how their collective actions coordinate for end-to-end paths (global issue).
Forwarding is a router-local action, moving packets from an input to an output port, typically in nanoseconds and implemented in hardware. Routing is a network-wide activity of determining paths from source to destination, taking place on a longer timescale (seconds) and often implemented in software. An analogy is crossing an intersection (forwarding) versus planning an entire road trip (routing).
The control plane, which determines datagram paths and handles network management, has two main approaches: traditional distributed routing algorithms and Software-Defined Networking (SDN). In traditional routing, algorithms run in individual routers and communicate to compute forwarding tables. In SDN, a remote controller computes and distributes forwarding tables to routers.
The Internet's network layer operates on a 'best effort service' model, meaning transmitted packets are not guaranteed delivery, bounded delay, or minimum bandwidth. While seemingly minimalist, this model has contributed significantly to the Internet's success.
The simplicity of the best effort service model facilitated widespread Internet deployment and management. Adequate bandwidth provisioning has enabled real-time services to work 'good enough.' The rise of distributed application-level infrastructure (like Netflix's content delivery) also compensates for potential network inconsistencies. Finally, TCP congestion control helps manage network congestion effectively.