How is data transmitted in an Ethernet network?

Data transmission on an Ethernet network is not a matter of an application simply dropping a “packet” onto the cable. Data is encapsulated layer by layer: the application protocol creates its message, TCP or UDP and IP information are added where required, the Ethernet interface places that data into a frame containing source MAC + destination MAC + payload + FCS, and the PHY transmits the bits using the selected physical medium. The receiving side reverses the process.

The key distinction is that Ethernet provides Layer-2 frame delivery across a hop; IP selects the path between networks; TCP can retransmit lost data when used; and the application protocol knows what the data means. Combining all of those jobs into one vague “Ethernet packet” model causes many troubleshooting mistakes.

Example: an HMI reads a temperature value from a PLC.

1. Application
The HMI creates the read request.
→ 2. TCP/UDP
Transport information is added as required by the protocol.
→ 3. IP
Source/destination IP and routing information are applied.
→ 4. Ethernet
The next-hop MAC is resolved and a frame is built.
→ 5. PHY
Bits are transmitted using the selected copper/fiber PHY.
10. HMI application
The PLC response reaches the display logic.
← 9. Upper layers
Ethernet payload → IP → TCP/UDP → application.
← 8. Receiving NIC
The frame/FCS is checked and the payload passed upward.
← 7. Switch
Forwards according to the destination MAC.
← 6. Physical link
The frame moves across the current hop.

1. Data starts at the application layer

When an HMI asks a PLC for a register, CIP attribute or process variable, the first object created is not an Ethernet frame. Modbus TCP, EtherNet/IP, OPC UA or another application/industrial protocol first creates its own message. Whether that protocol then uses TCP, UDP or direct Ethernet determines the next encapsulation steps.

That is why protocols that can operate over Ethernet and the Ethernet protocol are different topics. Ethernet provides the transport foundation; the upper protocol answers “which register?”, “which service?” or “which process data?”

2. What do TCP or UDP add?

When an IP-based application uses TCP, application bytes are carried in TCP segments and then in IP datagrams. RFC 9293 defines TCP reliability through loss detection and retransmission. [1]

Protocols using UDP do not receive TCP's connection, ACK and retransmission model. The application or industrial protocol can instead use cyclic refresh, watchdogs, sequence handling or its own recovery logic. Some EtherNet/IP I/O traffic, for example, uses UDP while explicit/parameter communication can use TCP.

3. Why is the IP destination different from the Ethernet destination?

The IP packet carries the final destination IP. The Ethernet frame carries the destination MAC for the current Layer-2 hop. If the target is on the same local subnet, the next hop is normally the destination host itself. If the target is on another IP network, the host sends the frame to the MAC address of the first-hop router/default gateway, not to the final remote device's MAC. Internet architecture describes this as choosing a local destination versus a first-hop router. [2]

SituationDestination in the IP packetDestination MAC in the first Ethernet frame
HMI and PLC on the same subnetPLC IP addressPLC MAC address
HMI and PLC on different subnetsStill the final PLC IP addressMAC address of the HMI-side default gateway/router

When the router forwards the packet into the next network, the old Layer-2 frame ends and the IP packet is placed into a new Ethernet frame for that hop. MAC addresses can therefore change at each routed hop while the IP endpoints continue to represent the original communicating hosts.

4. How is the destination MAC discovered?

On IPv4 Ethernet networks, a host that does not know the MAC address of the local next-hop IP can use ARP. RFC 826 defines the dynamic mapping between protocol addresses and 48-bit Ethernet addresses. [3]

An ARP request is commonly sent as a broadcast; the device or gateway owning the target IP replies with its MAC address. The mapping may then be cached. This is not the same process as a switch learning its MAC table. See What is a MAC address? for the address distinction.

Why are an ARP cache and a switch MAC table different?

TableWhere?StoresUsed for
ARP cacheHost, PLC, HMI, routerIPv4 address ↔ MAC addressFinding the next-hop MAC for a frame
Switch MAC/CAM tableEthernet switchMAC address ↔ switch port + VLANChoosing the output port for a frame

Confusing these two mechanisms leads to statements such as “the switch converts IP into MAC.” A normal Layer-2 switch forwards based on the destination MAC within the relevant VLAN/bridge domain.

5. How is the Ethernet frame created?

RFC 894 defines the standard encapsulation of IPv4 datagrams in Ethernet frames. [4] The Ethernet MAC carries upper-layer data in the Ethernet frame structure, including destination MAC, source MAC, Length/EtherType, MAC client data/pad and a 4-byte FCS.

IEEE 802.3 terminology defines the MAC frame from Destination Address through the FCS inclusive. [5] The FCS contains a 32-bit CRC calculated over the relevant frame fields. [6]

6. How does the PHY put data onto copper or fiber?

The PHY section of the Ethernet NIC converts the MAC bit stream into the signalling used by the selected Ethernet physical standard. There is no one universal “Ethernet encoding.” 10BASE-T, 100BASE-TX, 1000BASE-T, 10GBASE-T and optical PHYs use different line coding and modulation methods.

That is why a statement such as “Ethernet uses NRZ or 4B/5B” is only meaningful when tied to a particular PHY. The cable type, PHY, link speed and permitted distance must be evaluated together.

7. What does a switch do with the frame?

A modern switch examines the source MAC and VLAN of an arriving frame and learns which MAC is reachable on that port. It then looks up the destination MAC in its MAC/CAM table. Cisco's current switching documentation describes this behaviour explicitly. [7]

Destination-MAC stateSwitch action
Known on another portForward only out the associated port.
Known on the incoming portDo not send it back out that port; filter it.
Unknown unicastFlood to the other applicable ports in the same VLAN.
BroadcastFlood to the other applicable ports in the same VLAN/broadcast domain.
MulticastDepends on multicast learning/snooping and configuration; unknown multicast may be flooded.

So the sentence “a switch always sends a frame only to the destination port” is incomplete. Unknown-unicast flooding is a normal part of switching before the destination is learned. [7] See Ethernet switch vs. hub for details.

8. How does the receiving device accept the frame?

The receiving PHY recovers the bit stream and the MAC applies the destination-address and frame-validity rules. If the frame is intended for the interface, its FCS/CRC is checked. IEEE 802.3 uses a 32-bit CRC in the FCS for detecting transmission corruption. [6]

If the frame is valid, the Ethernet payload can be passed upward. An IPv4 payload goes to IP; TCP or UDP is processed next when present; and finally Modbus TCP, EtherNet/IP explicit messaging, HTTP or another application protocol consumes the data.

9. If the FCS is bad, does Ethernet request a retransmission?

No. A frame with an invalid FCS is normally discarded by the receiving Ethernet MAC. The receiver does not send an application-style acknowledgement saying “frame 27 was corrupt, please resend the same frame.” FCS is an error-detection mechanism, not an automatic end-to-end retransmission protocol.

This corrects the main contradiction in the older page: it says Ethernet has no frame acknowledgement, but also says a bad frame normally causes the receiver to send an error message requesting retransmission. TCP is the layer that can provide loss recovery when TCP is used. RFC 9293 explicitly defines TCP reliability through acknowledgements and retransmission. [1]

What happens when UDP or a real-time protocol is used?

UDP does not provide the same reliable byte-stream and retransmission behaviour as TCP. A UDP-based industrial protocol may instead refresh cyclic data, use watchdogs, sequence counters or connection timeouts. In a fast control loop, sending the newest process value in the next cycle can be more useful than retransmitting stale process data.

That is why “UDP is unreliable and therefore unusable in industry” is also wrong. Time-critical EtherNet/IP I/O can use UDP because the application protocol provides a timing model suited to cyclic control.

10. Does modern full-duplex Ethernet use CSMA/CD?

Shared half-duplex Ethernet historically used CSMA/CD to detect collisions and retry. Modern switched point-to-point full-duplex Ethernet allows both ends to transmit and receive simultaneously; there are no normal collisions and CSMA/CD is not used. Cisco's Ethernet documentation explicitly states that full-duplex mode has no collisions and does not require CSMA/CD. [8]

So describing a current PLC–switch–HMI link as “listen to the cable, send only when idle, back off after a collision” applies a historical half-duplex model to a modern full-duplex network.

Same-VLAN delivery vs. routed delivery

Same Layer 2 / VLANDifferent IP network / through a router
The sender can resolve the destination's MAC with ARP.The sender resolves the default-gateway/first-hop router MAC.
The switch forwards the frame toward the destination MAC.The first switch forwards to the router, and the router routes the IP packet.
The Layer-2 frame can remain inside one broadcast domain.At each routed hop, the old Ethernet frame ends and a new Layer-2 frame is created.
No router is required.The final destination IP remains the remote host while Layer-2 MAC addresses change per hop.

How does a VLAN affect delivery?

Switch forwarding operates in a VLAN/bridge-domain context as well as by MAC address. The same MAC value can exist in separate forwarding contexts in different VLANs. Broadcast and unknown-unicast flooding are normally constrained to the relevant VLAN.

Two PLCs may therefore have apparently compatible IP settings but still fail to communicate if their switch ports are isolated in different VLANs with no Layer-3 path. Conversely, routed communication between VLANs can work when a gateway and access policy are configured correctly. See Ethernet network security for the segmentation side.

Why doesn't EtherCAT fit this explanation exactly?

In a conventional TCP/IP Ethernet example, the destination receives a frame and passes the payload up the stack. EtherCAT process data uses a different model: as a frame passes through the network, SubDevices can read or write their assigned data on the fly. It should therefore not be explained as a normal switched TCP request/response exchange. See How does EtherCAT work?.

Where does total PLC-to-HMI latency come from?

Ethernet link speed is only one component of the total time. A process value reaching an HMI can be influenced by:

• PLC scan and communication-task time
• Application polling/RPI/update interval
• TCP/UDP processing and OS/firmware scheduling
• Frame serialization time and link rate
• Switch queuing, QoS and congestion
• Router/firewall processing when present
• HMI polling and display refresh interval

Moving a link from 100 Mb/s to 1 Gb/s therefore does not automatically make every HMI value update ten times faster. See Ethernet speeds for the difference between physical link rate and application-cycle performance.

“Link is up but there is no data” — what should you check?

1. Is the PHY link really up, with correct speed/duplex?
2. Are RX/TX, FCS/CRC, alignment or drop counters increasing?
3. Is the endpoint in the correct VLAN?
4. On the same subnet, does ARP learn the expected MAC?
5. Across subnets, are the default gateway and routing correct?
6. Does the switch MAC table learn the destination on the expected port?
7. Does the firewall/ACL pass the required TCP/UDP traffic?
8. Are the application protocol's port, connection, register, Assembly or device-name settings correct?

This order separates physical media, Layer 2, Layer 3 and the application. It avoids conclusions such as “no ping means the PLC is broken” or “a lit link LED proves the protocol is working.”

In one paragraph

Ethernet data transmission starts with an application message, which may be encapsulated in TCP/UDP and IP before a next-hop MAC is selected and an Ethernet frame is built. The PHY transmits the frame over the physical medium. A switch learns from source MAC addresses and forwards according to destination MAC; unknown-unicast and broadcast traffic may be flooded within the VLAN. The receiver uses FCS to detect corruption, but Ethernet does not request end-to-end retransmission by itself. When TCP is used, reliability and retransmission are handled above Ethernet. At every routed hop, a new Ethernet frame is created around the IP packet.

Related Ethernet guides

What is Ethernet?  •  How does Ethernet work?  •  Ethernet protocol  •  Ethernet frame structure  •  MAC address  •  Ethernet card / NIC  •