IEEE 802.3 (Ethernet) Header Formats and the Need for Padding

Before the header, the physical layer sends a 7-byte preamble (alternating 1010...) followed by a 1-byte SFD (Start Frame Delimiter, 10101011) to synchronise the receiver's clock. Two header formats then exist.

1. Ethernet II (DIX) frame — the format in use today

+-----------+------+--------+--------+--------+----------------+------+------+
| Preamble  | SFD  | Dest.  | Source | Type   |    DATA        |      | FCS  |
|  (7 B)    | (1 B)| MAC    | MAC    | (2 B)  | (46 - 1500 B)  |      | (4 B)|
|           |      | (6 B)  | (6 B)  |        |                |      |      |
+-----------+------+--------+--------+--------+----------------+------+------+
                                  | 0x0800 = IPv4   |
                                  | 0x0806 = ARP    |
                                  | 0x86DD = IPv6   |

• Destination MAC (6 B) — receiving NIC, or FF:FF:FF:FF:FF:FF for broadcast.
• Source MAC (6 B) — sender's NIC address.
• Type / EtherType (2 B) — identifies the encapsulated network-layer protocol. A value $\ge 0x0600$ (1536) is interpreted as a type.
• Data (46-1500 B) — the network-layer packet (padded to at least 46 B).
• FCS (4 B) — CRC-32 over destination address through data, used for error detection.

2. IEEE 802.3 frame

+---------+-----+--------+--------+--------+----------------+------+------+
|Preamble | SFD | DA     | SA     | Length | LLC / SNAP     | DATA | FCS  |
| (7 B)   |(1 B)| (6 B)  | (6 B)  | (2 B)  | (3 or 8 B hdr) |      | (4 B)|
+---------+-----+--------+--------+--------+----------------+------+------+
                                   Length = 46 ... 1500 (bytes of payload)
                                   LLC = DSAP (1 B) + SSAP (1 B) + Control (1 or 2 B)
                                   SNAP = Org code (3 B) + EtherType (2 B)

The 2-byte field after the source address is a Length (a value $< 1536$) giving the number of bytes in the payload, instead of a protocol type. The IEEE 802.2 LLC header (DSAP/SSAP/Control) identifies the upper-layer user; when no DSAP value exists for a protocol a SNAP header is added which carries the original EtherType — this is how 802.3 and Ethernet II stay compatible.

3. Why padding is required

• The minimum frame must be 64 bytes (measured from the destination address through the FCS):
  $$6 + 6 + 2 + 46 + 4 = 64 \text{ bytes}$$
  So the payload field must be at least 46 bytes. If the network-layer packet is shorter (e.g. a 28-byte ARP packet), the sender zero-pads the data field to 46 bytes; the FCS covers the padding, and the receiver strips the padding using the Length field after checking the FCS.

• Reason 1 — collision detection (slot time). Under CSMA/CD a station must still be transmitting when a collision from the farthest station can reach it, i.e. the frame transmission time must cover the round-trip propagation time:
  $$T_{frame} \ge 2\tau \quad\Rightarrow\quad 64 \text{ bytes} = 512 \text{ bit times at } 10 \text{ Mb/s} = \text{the Ethernet slot time}$$
  A shorter frame would finish before a collision signal arrived, producing an undetected late collision — the sender would think the frame was delivered while it was actually corrupted.

• Reason 2 — receiver detection. The minimum length guarantees enough signal energy/pulses for the receiver's squelch circuit to distinguish a real frame from line noise.

• Maximum size: 1518 bytes (6 + 6 + 2 + 1500 + 4), or 1522 bytes when a 4-byte 802.1Q VLAN tag is inserted.