14.1b Layer 2 (Data Link): Ethernet frames, MAC addresses, and switches¶

Welcome to Layer 2 of the OSI model — the Data Link layer. If Layer 1 (Physical) is about sending raw bits over a cable, Layer 2 is where those bits become organized, addressable data that can travel between devices on the same local network. For anyone starting in AI infrastructure, understanding Layer 2 is essential because it's the foundation for how servers, storage, and GPUs communicate inside a data center rack or cluster.

🌐 Context: Why Layer 2 Matters in AI Infrastructure¶

In an AI data center, you might have hundreds of servers connected to high-speed switches. Before any AI workload can start, those servers need to discover each other, send data reliably, and avoid collisions. Layer 2 handles all of this. It uses MAC addresses to identify devices and Ethernet frames to package data. Switches are the key hardware that forwards these frames intelligently.

⚙️ Ethernet Frames — The Data Package¶

An Ethernet frame is the structured container that carries data across a local network. Think of it like a shipping box with a label.

🧩 Frame Structure (Simplified)¶

• Preamble — A short signal to synchronize timing between sender and receiver.
• Destination MAC Address — The hardware address of the intended receiver.
• Source MAC Address — The hardware address of the sender.
• EtherType — Indicates what kind of data is inside (e.g., IPv4, IPv6, ARP).
• Payload — The actual data being sent (e.g., a piece of an AI training dataset).
• Frame Check Sequence (FCS) — A checksum to detect errors during transmission.

📏 Key Points¶

• Frames are variable in size, typically between 64 and 1518 bytes (standard Ethernet).
• In AI data centers, jumbo frames (up to 9000 bytes) are often used to reduce overhead when moving large datasets.
• Frames are not routable across different networks — they only work within the same broadcast domain (a single LAN segment).

🕵️ MAC Addresses — The Hardware Identity¶

A MAC (Media Access Control) address is a unique identifier burned into every network interface card (NIC) at the factory. It's like a serial number for your network hardware.

🔑 MAC Address Format¶

• 48 bits long, usually written as 12 hexadecimal characters.
• Example: 00:1A:2B:3C:4D:5E
• The first 6 hex digits represent the Organizationally Unique Identifier (OUI) — the manufacturer.
• The last 6 hex digits are the device-specific part.

🧠 Important Notes¶

• MAC addresses are flat — they don't have a hierarchy like IP addresses.
• They are permanent (though they can be spoofed or overridden in software).
• In AI infrastructure, engineers often use MAC addresses to:
• Identify which physical port a server is connected to.
• Configure port security on switches.
• Troubleshoot connectivity issues at the hardware level.

📊 Visual Representation: OSI Layer 2 Ethernet Frame Structure¶

This diagram maps Layer 2 Ethernet framing, illustrating header fields, MAC addressing, and the CRC integrity check.

🛠️ Switches — The Intelligent Forwarding Engine¶

A switch is a Layer 2 device that connects multiple devices (servers, storage, GPUs) on the same network. Unlike a hub (which blindly repeats data to all ports), a switch learns which MAC addresses are on which ports and forwards frames only where they need to go.

🔄 How a Switch Works¶

1. Learning — When a frame arrives, the switch records the source MAC address and the port it came from in a MAC address table.
2. Forwarding — The switch looks up the destination MAC address in its table. If found, it sends the frame only to that specific port.
3. Flooding — If the destination MAC is unknown, the switch sends the frame to all ports (except the one it came from).
4. Filtering — The switch drops frames that are meant for devices on the same port (no need to forward).

📊 Comparison: Hub vs. Switch vs. Router¶

🔗 Putting It All Together: A Simple Example¶

Imagine two AI training servers, Server A (MAC: AA:BB:CC:DD:EE:01) and Server B (MAC: AA:BB:CC:DD:EE:02), connected to the same switch.

1. Server A wants to send a small batch of training data to Server B.
2. It creates an Ethernet frame with:
3. Destination MAC: AA:BB:CC:DD:EE:02
4. Source MAC: AA:BB:CC:DD:EE:01
5. Payload: the training data chunk
6. The frame travels to the switch.
7. The switch checks its MAC address table and sees that MAC AA:BB:CC:DD:EE:02 is on Port 2.
8. The switch forwards the frame only to Port 2.
9. Server B receives the frame, checks the destination MAC, and processes the data.

This process happens millions of times per second in a busy AI cluster.

🧪 Practical Tips for New Engineers¶

• Check MAC addresses when troubleshooting connectivity — a mismatch can indicate a bad cable, wrong port, or NIC failure.
• Use jumbo frames carefully — all devices on the same Layer 2 segment must support the same MTU (Maximum Transmission Unit) size.
• Watch for MAC flooding — if a switch's MAC address table fills up, it may fall back to flooding all frames, causing performance degradation.
• Understand VLANs — in larger AI data centers, VLANs (Virtual LANs) are used to segment Layer 2 traffic logically, even on the same physical switch.

✅ Summary¶

• Layer 2 is the Data Link layer, responsible for reliable communication between directly connected devices.
• Ethernet frames are the structured data units that carry payloads across a LAN.
• MAC addresses are hardware-level identifiers that uniquely identify each network interface.
• Switches intelligently forward frames based on MAC addresses, enabling efficient communication within a network segment.

Mastering these basics will give you a solid foundation for understanding how AI workloads move data inside the data center — and how to keep that movement fast and reliable.