Networking Switch Fabric and Forwarding Tables

Chapter 4: Network Layer: The Data Plane

• This chapter focuses on the data plane of the network layer, focusing on how data is moved through a network.
• The chapter will cover network layer service models, forwarding versus routing, how a router works, addressing, and generalized forwarding.
• The implementation of the IP protocol and NAT in the Internet will also be discussed.
• The chapter will also cover generalized forwarding and SDN.

Goals

• Understanding principles behind network layer services, specifically the data plane.
• Identifying various network layer service models.
• Differentiating between forwarding and routing processes.
• Explaining the router's working mechanism.
• Discussing addressing schemes.
• Covering generalized forwarding methods involved in transferring data between routers.
• Providing implementation examples in the Internet.
• Explaining the IP protocol to understand the datagram format, fragmentation, IPv4 addressing, network address translation (NAT), and IPv6.
• Explaining various components involved in forwarding data within a router, including input ports, output ports, the switching fabric, and the routing processor.

Network Layer Services and Protocols

• The network layer is responsible for transporting segments from the sender to the receiver.
• The sender encapsulates segments into datagrams, which are passed to the link layer.
• The receiver delivers segments to the transport layer.
• This layer involves protocols relevant in every Internet device, such as hosts and routers.
• Routers examine header fields in IP datagrams to determine the appropriate output ports for forwarding data.

Two Key Network-Layer Functions

• Forwarding: The process of moving packets from a router's input to its appropriate output. This process gets the packet through a single interchange.
• Routing: The process of determining the route a packet takes from source to destination. Routing algorithms are used to make these decisions.
• Both forwarding and routing work together, with routing algorithms determining forwarding table entries in routers.

Two Control-Plane Approaches

• Traditional routing algorithms control how data flows. These algorithms are implemented in the router itself
• SDN (software-defined networking) uses a remote controller to control routing decisions. This controller computes and installs forwarding tables in routers.

Network Service Model

• Question: What service model for transporting datagrams from sender to receiver?
• Example: Guaranteed delivery, guaranteed delivery with low delay, in-order datagram delivery.
• Example Services for a flow of datagrams: guaranteed minimum bandwidth to the flow, constraints on changes in inter-packet spacing.

Network Layer Service Models (QoS Guarantees)

• Quality of Service (QoS) Guarantees
• Bandwidth
• Loss
• Order
• Timing
• Congestion
• Internet: best effort, no guarantees.
• ATM (Asynchronous Transfer Mode): Various service models with different QoS guarantees (CBR, VBR, ABR, UBR)

Network Layer: Data Plane and Control Plane

• Data plane: The forwarding function determines how a datagram arriving at a router's input port is forwarded to the appropriate output port. This function is local and performed at each router.
• Control plane: The routing function determines the route taken by datagrams among routers to go from source to destination. This function is network-wide rather than local.

IP Datagram Format

• 32 bits: IP datagram is structured as series of fields (version, header length, type of service, datagram length, etc...)

IP Fragmentation/Reassembly

• Network links have a maximum transmission unit (MTU) for datagrams.
• Datagrams larger than the MTU are fragmented into smaller datagrams.
• Datagrams are reassembled at the destination.
• Header bits in IP datagrams are used for identifying and ordering fragments.

Switching Fabrics

• Transferring packets from input to output links.
• Measured as a multiple of input/output line rate.
• Three types, Memory, Bus, and Crosbar.

Switching via Memory, Bus, and Interconnection Network

• Switching via memory is the first generation method, but limited by memory bandwidth.
• Switching via a bus is a better method but can experience contention, which limits bandwidth.
• Switching via an interconnection network (crossbar) overcomes bandwidth limitations by using multiple switching planes in parallel.

Input/Output Port Queueing

• Input port queueing refers to datagrams arriving at a router input port faster than the amount of data being forwarded by the router to the next link.
• Output port contention arises from output ports being slower than the input port(s).

How Much Buffering

• RFC 3439: An average rule of thumb to determine how much buffering is needed.
• More recent recommendations for buffering equal to RTT*C/N.

Router Architecture Overview

• Two key router functions:
• running routing algorithms/protocols
• forwarding datagrams.
• Four router components:
• input ports
• output ports
• switching fabric
• routing processor

Input Port Functions

• Physical layer (bit-level reception).
• Data link layer (e.g., Ethernet reception).
• Decentralized Switching.
• Lookup output port using a forwarding table in input port memory to forward destination based or generalized forwarding (SDNs).

Destination-Based Forwarding

• Uses forwarding tables to determine the output interface for a packet based on the destination address.

Longest Prefix Matching

• When looking for an entry in the forwarding table, use the longest address prefix that matches the destination address.

Content Addressable Memory (CAM)

• An approach to look up an address in forwarding table (fast lookup operation).

IPv4 Addressing and Subnetting

• IP addresses are used to identify hosts in networks.
• They are typically written in dotted decimal notation. Each part of the IP address is represented by a 8 bits number.
• Subnetting is the process of dividing a network into multiple subnetworks.
• Subnet masks are used to divide an IP address into network and host portions.
• Understanding how to determine network address, broadcast address, usable IP addresses is essential.

Classful Addressing

• Various class systems (A, B, C, D, E) had different structures separating IP addresses into network and host portions.

Classless Inter-Domain Routing (CIDR)

• More flexible than classful addressing.
• Subnet portion of an IP address can be of arbitrary length.
• Addresses are arranged as network portion followed by host portion.

DHCP (Dynamic Host Configuration Protocol)

• Automatically assigns IP addresses to hosts on a network.
• Supports network mobility, and simplifies administrator’s work.
• A four-step process: discover, offer, request, and acknowledge(ACK).