Network Layer Switching and Services

The network layer acts as the backbone of end-to-end communication across the global Internet, abstracting the underlying physical transmission media into a cohesive, logical routing structure. By managing logical addressing, packet switching, error reporting, and hardware translation, this layer ensures that data packets are successfully routed from source to destination across diverse physical networks.

1. Switching Paradigms: Circuit vs. Packet Switching

At the architectural level, transport paths across intermediate nodes must rely on switching mechanisms to route data. Computer networks historically evolved from circuit-switched systems to modern packet-switched systems.

text

               +-------------------------------------------+
               |             Switching Methods             |
               +--------------------+----------------------+
                                    |
                +-------------------+-------------------+
                |                                       |
     +----------v----------+                 +----------v----------+
     |  Circuit Switching  |                 |   Packet Switching  |
     +---------------------+                 +----------+----------+
                                                        |
                                        +---------------+---------------+
                                        |                               |
                             +----------v----------+         +----------v----------+
                             |    Connectionless   |         | Connection-Oriented |
                             |  (Datagram Network) |         |  (Virtual Circuit)  |
                             +---------------------+         +---------------------+

Circuit-Switched Networks

In circuit switching, a dedicated physical or logical path is established between the source and destination before transmission begins.

A classic example is the early analog telephone system.
The whole message is transmitted as an uninterrupted continuous stream; it is not divided into packets.
Resources along the path (bandwidth, switches) are exclusively reserved for the duration of the session, leading to low utilization efficiency during idle periods.

Packet-Switched Networks

In packet switching, messages are segmented into smaller, self-contained units called packets (or datagrams at the IP layer). These packets are individually routed through the network.

Packets may take different paths depending on network conditions.
Resources are allocated dynamically on a demand basis (statistical multiplexing).

Connectionless vs. Connection-Oriented Packet Services

Modern packet-switched network layers are implemented as either connectionless or connection-oriented services.

1. Connectionless Service (Datagram Networks)

In a connectionless packet-switched network, each packet is treated as an independent entity called a datagram.

Routing Decisons: The forwarding decision at each intermediate router is made solely based on the destination address carried in the packet header.
Pathing: Packets belonging to the same flow can take different physical routes, potentially arriving at the destination out-of-order.
Forwarding Process: Each router maintains a routing table that maps destination address ranges to output interfaces.

text

   Packet [SA|DA|Data] ---> [ Router ] ---> Output Interface 2 (Based on DA)

2. Connection-Oriented Service (Virtual-Circuit Networks)

In a connection-oriented packet-switched network, a logical path called a virtual circuit (VC) is established between the source and destination before any data packets are sent.

Routing Decisions: Forwarding decisions are based on a short identifier in the packet header called a label (Virtual Circuit Identifier, or VCI).
Setup Phase: A request packet is sent to reserve resources and populate translation tables in intermediate routers along the path. An acknowledgment packet completes the path setup.
Data Flow: Packets containing labels are forwarded rapidly. At each router, the incoming port and label are used to lookup the corresponding outgoing port and outgoing label.

text

    Router Virtual Circuit Table:
   +---------------+---------------+----------------+---------------+
   | Incoming Port | Incoming Label| Outgoing Port  | Outgoing Label|
   +---------------+---------------+----------------+---------------+
   |      1        |      L1       |       2        |      L2       |
   +---------------+---------------+----------------+---------------+

Teardown Phase: Once transmission is complete, a teardown signal is sent to release table entries and reserved resources.

Delay Characterization

In a connectionless network, the delay is dominated by packet processing and transmission times at each hop, but there is zero setup delay:

$\text{Delay}_{\text{Connectionless}} = \sum_{i=1}^{H} \left( T_{\text{proc}, i} + T_{\text{trans}, i} + T_{\text{prop}, i} \right)$

In connection-oriented virtual circuits, a setup time ( $\text{T}_{\text{setup}}$ ) and teardown time ( $\text{T}_{\text{teardown}}$ ) are introduced, but intermediate forwarding delays ( $T_{\text{forward}}$ ) are minimized due to simple label-switching lookups:

$\text{Delay}_{\text{Connection-Oriented}} = T_{\text{setup}} + \sum_{j=1}^{P} \left( T_{\text{trans}, j} + T_{\text{forward}, j} + T_{\text{prop}, j} \right) + T_{\text{teardown}}$

2. IPv4 Addressing and Binary Manipulations

An IP address is a 32-bit identifier that uniquely and universally defines the connection of a host or router interface to the global Internet.

Notations and Range

The total address space of IPv4 is:

$2^{32} = 4,294,967,296 \text{ addresses}$

To make the binary addresses readable, three notations are used:

Dotted-Decimal Notation: Four decimal values separated by dots (e.g., 129.11.11.239). Each decimal represents $8$ bits ( $1$ byte), ranging from $0$ to $255$ .
Binary Notation: 32 bits grouped into octets (e.g., 10000001 00001011 00001011 11101111).
Hexadecimal Notation: 8 hexadecimal digits often prefixed by 0x (e.g., 0x810B0BEF).

text

Binary:   10000001 . 00001011 . 00001011 . 11101111
            |          |          |          |
Decimal:   129    .    11    .    11    .   239
            |          |          |          |
Hex:       81     .    0B    .    0B    .   EF   ---> 0x810B0BEF

Logical Bitwise Operations on IP Addresses

Manipulating IPv4 addresses and masks requires three primary bitwise operations:

Bitwise NOT ( $\sim$ ): Inverts each bit of the target address.
Bitwise AND ( $\&$ ): Yields a $1$ only if both corresponding bits are $1$ . Used to extract the network address from an IP address and a subnet mask.
Bitwise OR ( $|$ ): Yields a $1$ if at least one corresponding bit is $1$ . Used alongside a negated mask to find the last address (broadcast) in a block.

Address Span Computation

To calculate the total number of addresses in a block spanning from a starting address ( $A_{\text{start}}$ ) to an ending address ( $A_{\text{end}}$ ), we interpret the addresses as base-256 integers:

$\text{Address Span} = (A_{\text{end}} - A_{\text{start}}) + 1$

3. Classful and Classless Addressing Schemes

Historically, IP addresses were structured using classful boundaries. Due to inefficient address allocation, this was replaced by classless addressing.

Classful Addressing

In classful addressing, the 32-bit space was divided into five distinct classes: A, B, C, D, and E. The class of an address can be identified by analyzing its leading bits:

text

Class A:  0------------------------------------  (0.0.0.0 to 127.255.255.255)
Class B:  10-----------------------------------  (128.0.0.0 to 191.255.255.255)
Class C:  110----------------------------------  (192.0.0.0 to 223.255.255.255)
Class D:  1110---------------------------------  (224.0.0.0 to 239.255.255.255) - Multicast
Class E:  1111---------------------------------  (240.0.0.0 to 255.255.255.255) - Reserved

Classful addresses are divided into a network identifier (netid) and a host identifier (hostid):

Class A: 8-bit netid, 24-bit hostid. Mask is 255.0.0.0. Accommodates up to $2^{24}-2 = 16,777,214$ hosts per block.
Class B: 16-bit netid, 16-bit hostid. Mask is 255.255.0.0. Accommodates $2^{16}-2 = 65,534$ hosts per block.

Subnetting in Classful Networks

To avoid wasting internal addresses, organizations divided their allocated blocks into multiple smaller, physical subnetworks. This introduces a three-level hierarchy (Netid $\rightarrow$ Subnetid $\rightarrow$ Hostid). The default class mask is modified by borrowing bits from the hostid field to create the subnet mask.

Classless Addressing (CIDR)

Classless Inter-Domain Routing (CIDR) replaced classful boundaries with variable-length blocks. In classless addressing, a block of addresses is defined by an IP address and a prefix length ( $n$ ).

Prefix: Identifies the network (equivalent to netid, $n$ bits).
Suffix: Identifies the host ( $32 - n$ bits).

Slash Notation

CIDR uses slash notation: IP_Address/n, where $n$ is the prefix length (e.g., 192.168.1.0/24).

Block Calculations

For a given CIDR address block defined as $A/n$ :

Number of Addresses ( $N$ ):

$N = 2^{32-n}$
First Address (Network Address): Set the rightmost $32-n$ bits of the address to $0$ .

$First Address = Address AND Mask$

Worked Example

Consider the address 110.23.120.14/20.

Prefix length: $n = 20$ .
Suffix length: $32 - 20 = 12$ .
Subnet Mask: The first $20$ bits are $1$ , and the remaining $12$ bits are $0$ : .

4. Network Address Translation (NAT)

As the availability of public IPv4 addresses declined, Network Address Translation (NAT) was introduced as a transition technology. It allows a private network to use a single public IP address (or a small pool of addresses) for external Internet communication.

text

  Private Network (Inside)                        Public Network (Outside)
  +----------------------+                        +----------------------+
  | Host A: 192.168.1.10 | ---\                  |                      |
  +----------------------+     |                  |                      |
                               +--> [NAT Router] ---> Global Server:     |
  +----------------------+     |    (Public IP:   | 8.8.8.8              |
  | Host B: 192.168.1.20 | ---/     203.0.113.5)  |                      |
  +----------------------+                        +----------------------+

Private IP Address Ranges

The Internet Assigned Numbers Authority (IANA) reserved three blocks of IPv4 address space for private networks:

10.0.0.0 to 10.255.255.255 (10.0.0.0/8)
172.16.0.0 to 172.31.255.255 (172.16.0.0/12)
192.168.0.0 to 192.168.255.255 (192.168.0.0/16)

Translation Table Architectures

1. IP-Only Translation (Static/Dynamic NAT)

In an IP-only NAT translation scheme, the translation table maps one private IP directly to one public IP.

Limitation: If the NAT router has only $1$ global public IP, only one private host can communicate with an external server at a time. If a pool of $M$ public addresses is available, a maximum of $M$ concurrent hosts can access the outside world.

2. Network Address Port Translation (NAPT)

To bypass the limitations of IP-only mapping, modern NAT routers use Network Address Port Translation (NAPT). NAPT maps transport layer identifiers (port numbers) alongside IP addresses, allowing thousands of private sessions to share a single public IP address simultaneously.

text

                  NAPT Translation Table Example
   +-----------------------+-------------+----------------------+-------------+
   | Private Source IP     | Private Port| Public Source IP     | Public Port |
   +-----------------------+-------------+----------------------+-------------+
   | 192.168.1.10          |    1024     | 203.0.113.5          |    40001    |
   | 192.168.1.20          |    1024     | 203.0.113.5          |    40002    |
   +-----------------------+-------------+----------------------+-------------+

When an outgoing packet is intercepted:

The NAT router replaces the private source IP with its public IP.
The router replaces the private source port with a unique public port assigned from its local pool.
The mapping is stored in the translation table.
For returning traffic, the destination port is matched against the public port in the table to forward the packet back to the correct internal private host.

5. Address Resolution Protocol (ARP)

Every device on a physical network is identified by a logical (IP) address and a physical (MAC) address. When a host needs to transmit an IP datagram, it must encapsulate it in a Layer 2 frame, which requires knowing the destination's physical address. The Address Resolution Protocol (ARP) dynamically maps a 32-bit logical IP address to a 48-bit physical MAC address.

text

    +-------------------------------------------------------+
    |                      ARP Packet                       |
    +---------------------------+---------------------------+
    | Hardware Type (16 bits)   | Protocol Type (16 bits)   |
    +---------------------------+---------------------------+
    | Hardware Size (8 bits)    | Protocol Size (8 bits)    |
    +---------------------------+---------------------------+
    |                   Opcode (16 bits)                    |
    +-------------------------------------------------------+
    |             Sender Hardware Address (MAC)             |
    +-------------------------------------------------------+
    |              Sender Protocol Address (IP)             |
    +-------------------------------------------------------+
    |             Target Hardware Address (MAC)             |
    +-------------------------------------------------------+
    |              Target Protocol Address (IP)             |
    +-------------------------------------------------------+

Operation: Request and Reply

text

  Host A (192.168.1.1)                                Host B (192.168.1.2)
       |                                                    |
       | ------ [ARP Request (Broadcast: Who is 1.2?)] ---->|
       |                                                    | (Saves A's MAC)
       |<------- [ARP Reply (Unicast: I am 1.2, MAC_B)] ----|

ARP Request (Broadcast): When Host A needs to resolve Host B's IP, it broadcasts an ARP request packet inside an Ethernet frame with the destination MAC address set to FF:FF:FF:FF:FF:FF. Every host on the local segment processes this frame.
ARP Reply (Unicast): Host B recognizes its IP address in the target protocol address field, updates its own ARP table with Host A's mapping, and returns a unicast ARP reply directly to Host A.
Encapsulation: The ARP frame is assigned an Ethernet Type field of 0x0806.

Four Operational Cases of ARP

ARP is used in four distinct networking scenarios:

Host-to-Host (Same Network): A host wants to send a packet to another host on the same local network. ARP resolves the target host's physical address.
Host-to-Router: A host wants to send a packet to a host on another network. ARP resolves the local gateway router's physical address.
Router-to-Router: A router wants to forward a packet to another intermediate router. ARP resolves the next-hop router's physical address.
Router-to-Host: The destination router receives a packet and needs to deliver it to the final destination host on its locally attached network. ARP resolves the destination host's physical address.

Proxy ARP

Proxy ARP is a technique where a router answers ARP requests on behalf of another device. This is often used to join two physical networks separated by a router without using subnetting, or to enable remote access VPN connections.

6. Internet Control Message Protocol (ICMP)

Because the IP protocol is connectionless and best-effort, it lacks built-in mechanism for reporting transmission errors. The Internet Control Message Protocol (ICMP) serves as the diagnostic and control companion to IP.

text

             +-----------------------+
             |     IP Header         | (Protocol field = 1)
             +-----------------------+
             |     ICMP Header       | (Type, Code, Checksum)
             +-----------------------+
             |     ICMP Data         | (Diagnostic payload)
             +-----------------------+

ICMP messages are encapsulated directly inside IP datagrams.

ICMP Message Categories

ICMP messages are categorized into Error-Reporting Messages and Query Messages.

1. Error-Reporting Messages

Destination Unreachable (Type 3): Sent when a packet cannot reach its destination.
- Code 0: Network unreachable (routing table issue).
- Code 1: Host unreachable (hardware connection issue).
- Code 2 & 3: Protocol/Port unreachable. These codes are generated only by the destination host.
Source Quench (Type 4): Sent to inform the source that a packet was discarded due to congestion. It acts as an basic congestion-control signal, advising the source to reduce its transmission rate.
Time Exceeded (Type 11):
- Code 0: Sent by a router when the Time-to-Live (TTL) field of a datagram reaches $0$ .
- Code 1: Sent by the destination host if reassembly fragments do not arrive within the expected timeout period.
Parameter Problem (Type 12): Sent when a router or host detects a corrupted header parameter.
Redirection (Type 5): Sent by a router to inform a local host that there is a more optimal next-hop router available on the local segment.

2. Query Messages

Echo Request (Type 8) and Echo Reply (Type 0): Used to test connection reachability. These queries are the foundation of the ping utility.
Timestamp Request (Type 13) and Timestamp Reply (Type 14): Used to compute round-trip times (RTT) and synchronize clock variations between two devices.

Checksum Calculation

Like IP, ICMP uses a 16-bit checksum to detect corruption in its header and data payload.

Mathematical Steps

Set the checksum field to 0x0000.
Segment the entire ICMP message into 16-bit words.
Perform a 1s complement addition on the words.
Take the 1s complement of the final sum.
Write this value into the ICMP checksum field.

Summary

Switching Paradigms: The network layer relies on either connectionless datagram switching (where routing is based on the destination IP address) or connection-oriented virtual-circuit switching (where routing is based on path labels established during a handshake).
IPv4 Address Structuring: Addresses are 32-bit logical identifiers. The legacy classful system (A, B, C, D, E) was replaced by the more flexible classless system (CIDR), which uses prefix/suffix slash notation (/n) to reduce address waste.
Network Address Translation (NAT): NAT helps conserve IPv4 addresses by mapping private address spaces to globally routable public IPs. NAPT dynamically maps transport port numbers to support many concurrent sessions on a single public IP.
Hardware Mapping via ARP: ARP translates 32-bit IP logical addresses to 48-bit physical MAC addresses. ARP requests are broadcasted to all local nodes, while ARP replies are unicasted back to the sender.
Diagnostics via ICMP: ICMP acts as a control protocol for IP, reporting delivery errors (like Destination Unreachable or Time Exceeded) and enabling connectivity tests (such as Echo Request/Reply in ping).

Assessed progress

Network Layer Switching and Services

Key Concepts

Packet Switching Paradigms

Classless Inter-Domain Routing (CIDR)

Network Address Translation (NAT)

Address Resolution Protocol (ARP)

Internet Control Message Protocol (ICMP)

Quick Check

Computer Networks

Computer Networks