6-1 - Network Layer 1 - Intro and Router

Why Network Layer?

  • Responsible for delivering packets from source host to destination host.
  • Determines which intermediate hops (routes) to take, managed by routing algorithms.
  • Routing can be distributed (traditional) or centralized (SDN).
  • Distinction between the control plane and data plane:
    • Control plane: Determines the end-to-end route for the packet.
    • Data plane: Forwards the packet to the appropriate next hop based on local decisions.

Protocols in the Network Layer

  • Only one primary protocol is used to ensure consistency—IP (Internet Protocol) with versions IPv4 and IPv6.
  • A packet at this layer is referred to as a datagram.

Data Plane vs Control Plane

  • Data Plane:
    • Handles the local function of forwarding datagrams.
    • Operates in real-time (nanoseconds) at the router level.
  • Control Plane:
    • Manages broader network decisions and routing based on global metrics.
    • Operates on a slower timeframe (milliseconds).

Router Architecture

  • Components:
    • Input ports
    • Output ports
    • High-speed switching fabric
    • Routing processors
  • Functions of input ports include line termination, lookup, forwarding, and queueing
    • Forwarding is based on destination IP address, facilitating efficient routing.

Output Port Functions

  • Buffering is essential when datagrams arrive faster than transmission capacity.
  • Implement a drop policy: Determine which datagrams to drop under buffer overflow.
  • Scheduling: Determine which queued datagram gets sent next.

Destination-Based Forwarding

  • Forwarding decisions are based solely on the destination IP address in the forwarding table.
  • Longest Prefix Matching:
    • When searching for the appropriate entry, use the longest match in the address prefix.
    • Example: For destination address decisions needing matching to the forwarding table entries.

Switching Fabrics

  • Function: Transfer packets from input links to the chosen output links.
  • Switching rate should ideally match or exceed input/output line rates (N times).

Types of Switching Fabrics

  1. Memory-based Switching: Traditional methods reliant on bus memory; limited by memory bandwidth.
  2. Bus Switching: Direct switching through a shared bus which can lead to contention and limited speed.
  3. Interconnection Networks: Use multiple small switches to optimize switching speed and capacity via parallelism.

Queuing and Scheduling

  • Input Port Queueing:

    • When switching fabric speed is exceeded by input rates, queuing delays occur.
    • HOL (Head-of-Line) Blocking: The first packet in a queue can block others from being processed.

  • Output Queueing:

    • Buffering is required to manage faster arrivals than link transmission capability.
    • Implement drop policies and scheduling disciplines to decide datagram transmission.

Buffer Management

  • Buffering Rules: RFC 3439 recommends buffer size be equal to the product of average RTT and link capacity.
  • Managerial Strategies:
    • Drop strategies: e.g., tail drop vs. priority drops.
    • Marking: Identify packets for congestion signaling.

Packet Scheduling Techniques

  • First Come First Served (FCFS): Queued packets are transmitted in the order of arrival.
  • Priority Scheduling: Packets are queued and transmitted based on class or priority.
  • Round Robin: Cycles through queues, serving one packet from each.
  • Weighted Fair Queuing (WFQ): Guarantees minimum bandwidth for different traffic classes based on assigned weights.

Acknowledgements

  • Slides adopted from Kurose's Computer Networking Slides.