[Computer Networking] 16. Ethernet, Switches, and VLANs

MAC Address Structure
======================

1A:2B:3C:4D:5E:6F  (hexadecimal notation)

- 48 bits = 6 bytes
- First 24 bits: OUI (Organizationally Unique Identifier)
  Example: 00:1A:2B = Ayecom Technology
- Last 24 bits: Unique number assigned by manufacturer
- Broadcast address: FF:FF:FF:FF:FF:FF

IP Address vs MAC Address:
  IP Address:  Network layer, logical, changes with location
  MAC Address: Link layer, physical, fixed (embedded in NIC)

ARP Operation
===============

Host A (192.168.1.10) sends data to Host B (192.168.1.20)

1. A's ARP table does not have B's MAC address

2. A broadcasts ARP request:
   "What is the MAC address for 192.168.1.20?"
   Destination MAC: FF:FF:FF:FF:FF:FF (broadcast)
   Source MAC: A's MAC address

3. All nodes in the subnet receive the ARP request
   Only B responds: "My MAC address is BB:BB:BB:BB:BB:BB"
   Destination MAC: A's MAC address (unicast)

4. A stores B's mapping in ARP table (TTL: typically 20 minutes)

ARP Table Example:
  IP Address      | MAC Address          | TTL
  ----------------+---------------------+------
  192.168.1.20    | BB:BB:BB:BB:BB:BB   | 13 min
  192.168.1.1     | AA:AA:AA:AA:AA:AA   | 8 min

Inter-Subnet Communication
============================

Host A (10.0.0.2) --> Host B (20.0.0.2)

Subnet 1              Router                Subnet 2
[A: 10.0.0.2]  ----  [10.0.0.1 | 20.0.0.1]  ----  [B: 20.0.0.2]
 MAC: AA              MAC: R1L  | MAC: R1R           MAC: BB

Step 1: A sends to router
  Source MAC: AA, Destination MAC: R1L
  Source IP: 10.0.0.2, Destination IP: 20.0.0.2

Step 2: Router sends to B
  Source MAC: R1R, Destination MAC: BB
  Source IP: 10.0.0.2, Destination IP: 20.0.0.2

  --> MAC addresses change per hop, IP addresses remain end-to-end

Ethernet Frame Format
=======================

+----------+----------+----------+------+----------+-----+
| Preamble | Dest MAC | Src MAC  | Type |   Data   | CRC |
| 8 bytes  | 6 bytes  | 6 bytes  | 2 B  | 46~1500  | 4 B |
+----------+----------+----------+------+----------+-----+

Field Descriptions:
  Preamble (8 bytes):
    - 7 bytes: 10101010 pattern (clock synchronization)
    - 1 byte: 10101011 (SFD, Start of Frame Delimiter)

  Dest MAC (6 bytes): Destination MAC address
  Src MAC (6 bytes):  Source MAC address

  Type (2 bytes): Upper layer protocol identifier
    - 0x0800: IPv4
    - 0x0806: ARP
    - 0x86DD: IPv6

  Data (46~1500 bytes): Payload
    - Minimum 46 bytes (padded if less)
    - Maximum 1500 bytes (MTU)

  CRC (4 bytes): Error detection (CRC-32)

Ethernet Characteristics
==========================

- Connectionless: No handshake
- Unreliable: No ACK/NAK, erroneous frames are simply discarded
  (Upper layer TCP handles retransmission)
- Uses CSMA/CD (half-duplex mode)
  (CSMA/CD not needed in full-duplex mode)

Ethernet Speed Evolution:
  10 Mbps    (10BASE-T)      1990s
  100 Mbps   (Fast Ethernet)  1995
  1 Gbps     (Gigabit)        1999
  10 Gbps    (10GbE)          2002
  40/100 Gbps                 2010
  400 Gbps                    2017

Switch Operation Overview
===========================

     Port1    Port2    Port3    Port4
      |        |        |        |
   +--+--------+--------+--------+--+
   |       Link-Layer Switch         |
   +--+--------+--------+--------+--+
      |        |        |        |
    [A]      [B]      [C]      [D]

A sends frame to C:
  1. Switch receives frame on Port 1
  2. Destination MAC = C's MAC
  3. Switch table lookup: C is on Port 3
  4. Forward frame only to Port 3
  --> B and D do not receive this frame

Self-Learning Algorithm
=========================

When frame arrives on port x with source MAC = AA:
  --> Record in switch table: MAC AA is on port x (set TTL)

Switch Table Building Process:
  Time 0: Table is empty

  Time 1: A (Port 1) sends frame to C
    Table: A -> Port 1
    C's location unknown --> Flooding (send to all ports except Port 1)

  Time 2: C (Port 3) replies to A
    Table: A -> Port 1, C -> Port 3
    A's location is known --> Forward only to Port 1

  Time 3: B (Port 2) sends to A
    Table: A -> Port 1, C -> Port 3, B -> Port 2
    A's location is known --> Forward only to Port 1

Spanning Tree Example
=======================

Topology with loops:

  [SW1] ---- [SW2]
    |    \    / |
    |     \ /   |
    |      X    |
    |     / \   |
    |    /    \ |
  [SW3] ---- [SW4]

After STP applied (some links deactivated):

  [SW1] ---- [SW2]
    |              |
    |              |
  [SW3]       [SW4]

  Deactivated links: SW1-SW4, SW2-SW3, SW3-SW4
  --> Loop-free tree structure formed

Switch vs Router Comparison
==============================

Item               | Switch (L2)          | Router (L3)
-------------------+----------------------+--------------------
Operating Layer    | Link Layer (L2)      | Network Layer (L3)
Address Used       | MAC Address          | IP Address
Table              | Switch Table         | Forwarding Table
Plug and Play      | Self-learning (yes)  | Configuration needed (partial)
Loop Handling      | STP required         | TTL natural elimination
Broadcast          | Propagates everywhere| Can be blocked
Scalability        | Limited at large scale| Hierarchical scaling possible

When VLANs Are Needed
========================

Physically one switch:
  Ports 1-4:   Marketing team
  Ports 5-8:   Development team
  Ports 9-12:  Management

Without VLAN: All broadcasts propagate to all ports
  --> Security issues, performance degradation

With VLAN:
  VLAN 10 (Marketing): Ports 1-4
  VLAN 20 (Development): Ports 5-8
  VLAN 30 (Management): Ports 9-12

  --> Each VLAN is an independent broadcast domain
  --> Inter-VLAN communication only through a router

VLAN Trunking
===============

[Switch 1]                        [Switch 2]
VLAN 10: Ports 1,2     Trunk      VLAN 10: Ports 1,2
VLAN 20: Ports 3,4  <=========>   VLAN 20: Ports 3,4

Trunk port: Carries frames from multiple VLANs over a single link

802.1Q Tag (4 bytes):
+--------+-----+-------+---------+
| TPID   | PRI | CFI   | VLAN ID |
| 0x8100 | 3b  | 1b    | 12 bits |
+--------+-----+-------+---------+

VLAN ID: 0~4095 (4096 VLANs possible)

Operation: Add 802.1Q tag when sending frame over trunk
           Receiver checks tag and forwards to appropriate VLAN port

MPLS Operation
================

Normal IP Routing:
  Each router performs longest prefix matching based on IP address (slow)

MPLS Routing:
  Ingress router assigns label --> Intermediate routers only reference label (fast)

Packet Structure:
  [L2 Header][MPLS Label][IP Header][Data]

MPLS Header (4 bytes):
  +-------+-----+---+-----+
  | Label | Exp | S | TTL |
  | 20bit | 3b  | 1b| 8b  |
  +-------+-----+---+-----+

Advantages:
  - Faster forwarding than IP routing
  - Traffic engineering possible
  - Easy VPN configuration

Data Center Network Topology
===============================

Traditional Hierarchical Structure:

          [Core Switch]
         /      |      \
    [Aggregation] [Aggregation] [Aggregation]
    /    \     /    \     /    \
 [ToR]  [ToR] [ToR] [ToR] [ToR] [ToR]
  |||    |||   |||   |||   |||   |||
 Servers Servers Servers Servers Servers Servers

ToR: Top-of-Rack switch (located on top of rack)
Aggregation: Aggregation switch
Core: Core switch

Problem: Bandwidth bottleneck toward upper layers

Fat-Tree Topology
====================

         [Core Switches]
        / | | | | | | | \
   [Agg1]   [Agg2]   [Agg3]   [Agg4]
   / | \    / | \    / | \    / | \
 [ToR] [ToR] [ToR] [ToR] [ToR] [ToR]

Features:
  - Equal bandwidth at all layers
  - Load balancing through multiple paths
  - Can use inexpensive commodity switches
  - Maximizes bisection bandwidth between servers

Data Center Load Balancing
============================

External Request --> [Load Balancer] --> Server 1
                                    --> Server 2
                                    --> Server 3
                                    --> Server 4

Load Balancer Roles:
  - Exposes a single public IP to the outside
  - Distributes incoming requests to internal servers
  - Monitors server health (health checks)
  - L4 (TCP port-based) or L7 (HTTP-based) distribution