4.2 What's Inside a Router? (Part 1)

Router Components: A router consists of four main components:
- Input Ports: Perform physical and link layer functions, and most importantly, perform the lookup function to determine the output port.
- Output Ports: Store packets received from the switching fabric and transmit them on the outgoing link.
- Switching Fabric: The heart of the router that connects input ports to output ports.
- Routing Processor: Executes routing protocols, maintains routing tables, and performs network management functions.
Control Plane vs. Data Plane:
- Data Plane: Operates at nanosecond timescales; it is the hardware-implemented forwarding path that moves packets from input to output.
- Control Plane: Operates at millisecond timescales; it is the software-driven logic (e.g., OSPF, BGP, or SDN controllers) that determines the paths packets take.

Line Termination: The physical layer interface receiving the incoming signal.
Data Link Processing: Handles protocol-specific framing (e.g., Ethernet, PPP).
Lookup and Forwarding:
- To achieve wire-speed forwarding, lookup must be performed in hardware.
- Ternary Content Addressable Memory (TCAM): Specialized hardware used to perform lookups in constant time, $O(1)$ , regardless of table size.
Input Queuing: Occurs if the switching fabric is slower than the combined speed of input ports.
- Head-of-the-Line (HOL) Blocking: A queued packet at the front of the line prevents following packets from moving forward, even if their output ports are available.

Destination-Based Forwarding: Unlike traditional switching, routers use IP address ranges to manage millions of potential destinations efficiently.
Longest Prefix Matching (LPM): When searching the forwarding table for a destination address, the router uses the entry with the longest matching prefix. This allows for hierarchical addressing and more specific routing rules.

Switching via Memory: The earliest method where the CPU controlled the transfer. Speed is limited by memory bandwidth (two crossings per packet).
Switching via Bus: Packets are transferred directly from input to output via a shared bus. Only one packet can cross the bus at a time; speed is limited by bus capacity.
Switching via Interconnection Network: Uses a complex fabric like a 2D mesh or Crossbar to allow multiple packets to be switched in parallel.
- CLOS Networks: Multi-stage interconnection networks that provide non-blocking paths and scalability for massive backbone routers.

Buffering: Required when packets arrive from the fabric faster than the output link speed.
- Rule of Thumb: The amount of buffering ( $B$ ) should be equal to the average Round Trip Time ( $RTT$ ) multiplied by the link capacity ( $C$ ), expressed as $B = RTT imes C$ .
- Recent Research: Suggests that for $N$ flows, the buffer size can be reduced to $B = rac{RTT imes C}{\sqrt{N}}$ .
Scheduling Disciplines: Determines which packet to send next:
- First-In-First-Out (FIFO)
- Priority Queuing: Segregates traffic into classes (e.g., Voice vs. Data).
- Weighted Fair Queuing (WFQ): Provides guaranteed bandwidth for different traffic flows.
Packet Discard Policy: If the buffer is full, the router must decide which packet to drop (e.g., Tail Drop, Random Early Detection).

Switching Capacity: Throughput is maximized by hardware parallelism. High-end routers can reach capacities in the tens of Terabits per second (Tbps).
Fabric Plane Parallelism: Modern routers use multiple switching fabric cards in parallel to ensure redundancy and higher aggregate throughput.