Fundamentals of Ethernet LANs

Most enterprise computer networks can be separated into two general types of technology: local-area networks (LAN) and wide-area networks (WAN).

LANs typically connect nearby devices: devices in the same room, in the same building, or in a campus of buildings.

In contrast, WANs connect devices that are typically relatively far apart.

Together, LANs and WANs create a complete enterprise computer network, working together to do the job of a computer network: delivering data from one device to another

Many types of LANs have existed over the years, but today’s networks use two general types of LANs: Ethernet LANs and wireless LANs.

Ethernet LANs happen to use cables for the links between nodes, and because many types of cables use copper wires, Ethernet LANs are often called wired LANs.

In comparison, wireless LANs do not use wires or cables, instead using radio waves for the links between nodes.

The term Ethernet refers to a family of LAN standards that together define the physical and data link layers of the world’s most popular wired LAN technology.

The standards, defined by the Institute of Electrical and Electronics Engineers (IEEE), define the cabling, the connectors on the ends of the cables, the protocol rules, and everything else required to create an Ethernet LAN.

To begin, first think about a small office/home office (SOHO) LAN today, specifically a LAN that uses only Ethernet LAN technology.

First, the LAN needs a device called an Ethernet LAN switch, which provides many physical ports into which cables can be connected.

An Ethernet uses Ethernet cables, which is a general reference to any cable that conforms to any of several Ethernet standards.

The LAN uses Ethernet cables to connect different Ethernet devices or nodes to one of the switch’s Ethernet ports.

Below diagram shows a drawing of a SOHO Ethernet LAN. The figure shows a single LAN switch, five cables, and five other Ethernet nodes: three PCs, a printer, and one network device called a router. (The router connects the LAN to the WAN, in this case to the Internet.)

Although above diagram shows a simple Ethernet LAN, many SOHO Ethernet LANs today combine the router and switch into a single device.

Vendors sell consumer-grade integrated networking devices that work as a router and Ethernet switch, as well as doing other functions.

These devices typically have “router” on the packaging, but many models also have four-port or eight-port Ethernet LAN switch ports built in to the device

Typical SOHO LANs today also support wireless LAN connections.

Ethernet defines wired LAN technology only; in other words, Ethernet LANs use cables.

However, you can build one LAN that uses both Ethernet LAN technology as well as wireless LAN technology, which is also defined by the IEEE.

Wireless LANs, defined by the IEEE using standards that begin with 802.11, use radio waves to send the bits from one node to the next.

Most wireless LANs rely on yet another networking device: a wireless LAN access point (AP).

The AP acts somewhat like an Ethernet switch, in that all the wireless LAN nodes communicate with the Ethernet switch by sending and receiving data with the wireless AP.

Of course, as a wireless device, the AP does not need Ethernet ports for cables, other than for a single Ethernet link to connect the AP to the Ethernet LAN, as shown below.

Note that this diagram shows the router, Ethernet switch, and wireless LAN access point as three separate devices so that you can better understand the different roles.

However, most SOHO networks today would use a single device, often labeled as a “wireless router,” that does all these functions.

Enterprise networks have similar needs compared to a SOHO network, but on a much larger scale.

For example, enterprise Ethernet LANs begin with LAN switches installed in a wiring closet behind a locked door on each floor of a building.

The electricians install the Ethernet cabling from that wiring closet to cubicles and conference rooms where devices might need to connect to the LAN.

At the same time, most enterprises also support wireless LANs in the same space, to allow people to roam around and still work and to support a growing number of devices that do not have an Ethernet LAN interface.

Below is a conceptual view of a typical enterprise LAN in a three-story building.

Each floor has an Ethernet LAN switch and a wireless LAN AP.

To allow communication between floors, each per-floor switch connects to one centralized distribution switch.

For example, PC3 can send data to PC2, but it would first flow through switch SW3 to the first floor to the distribution switch (SWD) and then back up through switch SW2 on the second floor.

The figure also shows the typical way to connect a LAN to a WAN using a router. LAN switches and wireless access points work to create the LAN itself. Routers connect to both the LAN and the WAN.

To connect to the LAN, the router simply uses an Ethernet LAN interface and an Ethernet cable, as shown on the lower right of the diagram above.

The term Ethernet refers to an entire family of standards.

Some standards define the specifics of how to send data over a particular type of cabling, and at a particular speed.

Other standards define protocols, or rules, that the Ethernet nodes must follow to be a part of an Ethernet LAN.

All these Ethernet standards come from the IEEE and include the number 802.3 as the beginning part of the standard name.

Ethernet supports a large variety of options for physical Ethernet links given its long history over the last 40 or so years.

Today, Ethernet includes many standards for different kinds of optical and copper cabling, and for speeds from 10 megabits per second (Mbps) up to 100 gigabits per second (Gbps).

The standards also differ as far as the types of cabling and the allowed length of the cabling.

The most fundamental cabling choice has to do with the materials used inside the cable for the physical transmission of bits: either copper wires or glass fibers.

The use of unshielded twisted-pair (UTP) cabling saves money compared to optical fibers, with Ethernet nodes using the wires inside the cable to send data over electrical circuits.

Fiber-optic cabling, the more expensive alternative, allows Ethernet nodes to send light over glass fibers in the center of the cable.

Although more expensive, optical cables typically allow longer cabling distances between nodes.

To be ready to choose the products to purchase for a new Ethernet LAN, a network engineer must know the names and features of the different Ethernet standards supported in Ethernet products. The IEEE defines Ethernet physical layer standards using a couple of naming conventions.

The formal name begins with 802.3 followed by some suffix letters.

The IEEE also uses more meaningful shortcut names that identify the speed, as well as a clue about whether the cabling is UTP (with a suffix that includes T) or fiber (with a suffix that includes X).

The table below lists a few Ethernet physical layer standards.

First, the table lists enough names so that you get a sense of the IEEE naming conventions.

It also lists the four most common standards that use UTP cabling, because this book’s discussion of Ethernet focuses mainly on the UTP options.

While the physical layer standards focus on sending bits over a cable, the Ethernet data-link protocols focus on sending an Ethernet frame from source to destination Ethernet node. From a data-link perspective, nodes build and forward frames. The term frame specifically refers to the header and trailer of a data-link protocol, plus the data encapsulated inside that header and trailer.

The diagram below shows an example of the process. In this case, PC1 sends an Ethernet frame to PC3. The frame travels over a UTP link to Ethernet switch SW1, then over fiber links to Ethernet switches SW2 and SW3, and finally over another UTP link to PC3. Note that the bits actually travel at four different speeds in this example: 10 Mbps, 1 Gbps, 10 Gbps, and 100 Mbps, respectively.

So, what is an Ethernet LAN? It is a combination of user devices, LAN switches, and different kinds of cabling. Each link can use different types of cables, at different speeds. However, they all work together to deliver Ethernet frames from the one device on the LAN to some other device.

Before the Ethernet network as a whole can send Ethernet frames between user devices, each node must be ready and able to send data over an individual physical link.

The three most commonly used Ethernet standards: 10BASE-T (Ethernet), 100BASE-T (Fast Ethernet, or FE), and 1000BASE-T (Gigabit Ethernet, or GE) which respectively uses the cables for 10-Mbps, 100-Mbps, and 1000-Mbps Ethernet.

While it is true that Ethernet sends data over UTP cables, the physical means to send the data uses electricity that flows over the wires inside the UTP cable.

To better understand how Ethernet sends data using electricity, break the idea down into two parts: how to create an electrical circuit and then how to make that electrical signal communicate 1s and 0s.

First, to create one electrical circuit, Ethernet defines how to use the two wires inside a single twisted pair of wires, as shown on the diagram below.

The diagram does not show a UTP cable between two nodes, but instead shows two individual wires that are inside the UTP cable.

An electrical circuit requires a complete loop, so the two nodes, using circuitry on their Ethernet ports, connect the wires in one pair to complete a loop, allowing electricity to flow.

Note that in an actual UTP cable, the wires will be twisted together, instead of being parallel.

The twisting helps solve some important physical transmission issues.

When electrical current passes over any wire, it creates electromagnetic interference (EMI) that interferes with the electrical signals in nearby wires, including the wires in the same cable. (EMI between wire pairs in the same cable is called crosstalk.) 

Twisting the wire pairs together helps cancel out most of the EMI, so most networking physical links that use copper wires use twisted pairs.

The term Ethernet link refers to any physical cable between two Ethernet nodes.

To learn about how a UTP Ethernet link works, it helps to break down the physical link into those basic pieces, as shown below: the cable itself, the connectors on the ends of the cable, and the matching ports on the devices into which the connectors will be inserted.

First, think about the UTP cable itself. The cable holds some copper wires, grouped as twisted pairs.

The 10BASE-T and 100BASE-T standards require two pairs of wires, while the 1000BASE-T standard requires four pairs.

Each wire has a color-coded plastic coating, with the wires in a pair having a color scheme.

For example, for the blue wire pair, one wire’s coating is all blue, while the other wire’s coating is blue-and-white striped.

Many Ethernet UTP cables use an RJ-45 connector on both ends.

The RJ-45 connector has eight physical locations into which the eight wires in the cable can be inserted, called pin positions, or simply pins. 

These pins create a place where the ends of the copper wires can touch the electronics inside the nodes at the end of the physical link so that electricity can flow.

To complete the physical link, the nodes each need an RJ-45 Ethernet port that matches the RJ-45 connectors on the cable so that the connectors on the ends of the cable can connect to each node.

Computers often include this RJ-45 Ethernet port as part of a network interface card (NIC), which can be an expansion card on the PC or can be built in to the system itself. Switches typically have many RJ-45 ports because switches give user devices a place to connect to the Ethernet LAN.

The diagram above shows a connector on the left and ports on the right. The left shows the eight pin positions in the end of the RJ-45 connector. The upper right shows an Ethernet NIC that is not yet installed in a computer.

The lower-right part of the figure shows the side of a Cisco 2960 switch, with multiple RJ-45 ports, allowing multiple devices to easily connect to the Ethernet network.

Finally, while RJ-45 connectors with UTP cabling can be common, Cisco LAN switches often support other types of connectors as well. When you buy one of the many models of Cisco switches, you need to think about the mix and numbers of each type of physical ports you want on the switch.

To give its customers flexibility as to the type of Ethernet links, even after the customer has bought the switch, Cisco switches include some physical ports whose port hardware (the transceiver) can be changed later, after you purchase the switch.

For example, below shows a photo of a Cisco switch with one of the swappable transceivers. In this case, the figure shows an enhanced small form-factor pluggable (SFP+) transceiver, which runs at 10 Gbps, just outside two SFP+ slots on a Cisco 3560CX switch. The SFP+ itself is the silver colored part below the switch, with a black cable connected to it.

Straight-Through Cable Pinout

10BASE-T and 100BASE-T use two pairs of wires in a UTP cable, one for each direction, as shown in diagram below.

The figure shows four wires, all of which sit inside a single UTP cable that connects a PC and a LAN switch. In this example, the PC on the left transmits using the top pair, and the switch on the right transmits using the bottom pair.

For correct transmission over the link, the wires in the UTP cable must be connected to the correct pin positions in the RJ-45 connectors.

For example, in the diagram above, the transmitter on the PC on the left must know the pin positions of the two wires it should use to transmit. Those two wires must be connected to the correct pins in the RJ-45 connector on the switch, so that the switch’s receiver logic can use the correct wires.

To understand the wiring of the cable—which wires need to be in which pin positions on both ends of the cable—you need to first understand how the NICs and switches work.

As a rule, Ethernet NIC transmitters use the pair connected to pins 1 and 2; the NIC receivers use a pair of wires at pin positions 3 and 6. LAN switches, knowing those facts about what Ethernet NICs do, do the opposite: Their receivers use the wire pair at pins 1 and 2, and their transmitters use the wire pair at pins 3 and 6.

To allow a PC NIC to communicate with a switch, the UTP cable must also use a straight-through cable pinout.

The term pinout refers to the wiring of which color wire is placed in each of the eight numbered pin positions in the RJ-45 connector.

An Ethernet straight-through cable connects the wire at pin 1 on one end of the cable to pin 1 at the other end of the cable; the wire at pin 2 needs to connect to pin 2 on the other end of the cable; pin 3 on one end connects to pin 3 on the other, and so on.

Also, it uses the wires in one wire pair at pins 1 and 2, and another pair at pins 3 and 6.

The diagram below shows one final perspective on the straight-through cable pinout.

In this case, PC Larry connects to a LAN switch.

Note that the diagram again does not show the UTP cable, but instead shows the wires that sit inside the cable, to emphasize the idea of wire pairs and pins.

Crossover Cable Pinout

A straight-through cable works correctly when the nodes use opposite pairs for transmitting data.

However, when two like devices connect to an Ethernet link, they both transmit on the same pins.

In that case, you then need another type of cabling pinout called a crossover cable.

The crossover cable pinout crosses the pair at the transmit pins on each device to the receive pins on the opposite device.

This concept is much clearer with a diagram  below. 

The figure shows what happens on a link between two switches. The two switches both transmit on the pair at pins 3 and 6, and they both receive on the pair at pins 1 and 2. So, the cable must connect a pair at pins 3 and 6 on each side to pins 1 and 2 on the other side, connecting to the other node’s receiver logic. The top of the figure shows the literal pinouts, and the bottom half shows a conceptual diagram.

Choosing the Right Cable Pinouts

• Crossover cable: If the endpoints transmit on the same pin pair
• Straight-through cable: If the endpoints transmit on different pin pairs

The diagram below shows a LAN in a single building. In this case, several straight-through cables are used to connect PCs to switches. In addition, the cables connecting the switches require crossover cables.

NOTE: If you have some experience with installing LANs, you might be thinking that you have used the wrong cable before (straight-through or crossover) but the cable worked. Cisco switches have a feature called auto-mdix that notices when the wrong cable is used and automatically changes its logic to make the link work.

However, for the exams, be ready to identify whether the correct cable is shown in the diagrams.

1000BASE-T (Gigabit Ethernet) differs from 10BASE-T and 100BASE-T as far as the cabling and pinouts.

First, 1000BASE-T requires four wire pairs.

Second, it uses more advanced electronics that allow both ends to transmit and receive simultaneously on each wire pair.

However, the wiring pinouts for 1000BASE-T work almost identically to the earlier standards, adding details for the additional two pairs.

The straight-through cable connects each pin with the same numbered pin on the other side, but it does so for all eight pins—pin 1 to pin 1, pin 2 to pin 2, up through pin 8.

It keeps one pair at pins 1 and 2 and another at pins 3 and 6, just like in the earlier wiring. It adds a pair at pins 4 and 5 and the final pair at pins 7 and 8 as shown in the diagram below.

The Gigabit Ethernet crossover cable crosses the same two-wire pairs as the crossover cable for the other types of Ethernet (the pairs at pins 1,2 and 3,6). It also crosses the two new pairs as well (the pair at pins 4,5 with the pair at pins 7,8).

Although physical layer standards vary quite a bit, other parts of the Ethernet standards work the same way, regardless of the type of physical Ethernet link. 

One of the most significant strengths of the Ethernet family of protocols is that these protocols use the same data-link standard.

The Ethernet data-link protocol defines the Ethernet frame: an Ethernet header at the front, the encapsulated data in the middle, and an Ethernet trailer at the end as shown in the diagrams below with the description of each field.

The source and destination Ethernet address fields play a huge role in how Ethernet LANs work.

The general idea for each is relatively simple: The sending node puts its own address in the source address field and the intended Ethernet destination device’s address in the destination address field.

The sender transmits the frame, expecting that the Ethernet LAN, as a whole, will deliver the frame to that correct destination.

Ethernet addresses, also called Media Access Control (MAC) addresses, are 6-byte-long (48-bit-long) binary numbers.

MAC addresses as 12-digit hexadecimal numbers.

Cisco devices typically add some periods to the number for easier readability as well; for example, a Cisco switch might list a MAC address as 0000.0C12.3456.

Most MAC addresses represent a single NIC or other Ethernet port, so these addresses are often called a unicast Ethernet address.

The term unicast is simply a formal way to refer to the fact that the address represents one interface to the Ethernet LAN. (This term also contrasts with two other types of Ethernet addresses, broadcast and multicast, which will be defined later.)

The entire idea of sending data to a destination unicast MAC address works well, but it works only if all the unicast MAC addresses are unique.

If two NICs tried to use the same MAC address, there could be confusion.  

If two PCs on the same Ethernet tried to use the same MAC address, to which PC should frames sent to that MAC address be delivered?

Ethernet solves this problem using an administrative process so that, at the time of manufacture, all Ethernet devices are assigned a universally unique MAC address.

Before a manufacturer can build Ethernet products, it must ask the IEEE to assign the manufacturer a universally unique 3-byte code, called the organizationally unique identifier (OUI).

The manufacturer agrees to give all NICs (and other Ethernet products) a MAC address that begins with its assigned 3-byte OUI.

The manufacturer also assigns a unique value for the last 3 bytes, a number that manufacturer has never used with that OUI. As a result, the MAC address of every device in the universe is unique.

The diagram below shows the structure of the unicast MAC address, with the OUI.

Ethernet addresses go by many names:

• LAN address
• Ethernet address
• hardware address
• burned-in address (BIA)
• physical address
• universal address
• MAC address.

The term burned-in address (BIA) refers to the idea that a permanent MAC address has been encoded (burned into) the ROM chip on the NIC.

As another example, the IEEE uses the term universal address to emphasize the fact that the address assigned to a NIC by a manufacturer should be unique among all MAC addresses in the universe.

Broadcast and Multicast Address

In addition to unicast addresses, Ethernet also uses group addresses.

Group addresses identify more than one LAN interface card. A frame sent to a group address might be delivered to a small set of devices on the LAN, or even to all devices on the LAN.

In fact, the IEEE defines two general categories of group addresses for Ethernet:

• Broadcast address: Frames sent to this address should be delivered to all devices on the Ethernet LAN. It has a value of FFFF.FFFF.FFFF.
• Multicast addresses: Frames sent to a multicast Ethernet address will be copied and forwarded to a subset of the devices on the LAN that volunteers to receive frames sent to a specific multicast address.

The diagram below summarizes most of the details about MAC addresses.

While the Ethernet header’s address fields play an important and more obvious role in Ethernet LANs, the Ethernet Type field plays a much less obvious role.

The Ethernet Type field, or EtherType, sits in the Ethernet data link layer header, but its purpose is to directly help the network processing on routers and hosts.

Basically, the Type field identifies the type of network layer (Layer 3) packet that sits inside the Ethernet frame.

First, think about what sits inside the data part of the Ethernet frame shown in the diagram below.

Typically, it holds the network layer packet created by the network layer protocol on some device in the network.

Over the years, those protocols have included IBM Systems Network Architecture (SNA), Novell NetWare, Digital Equipment Corporation’s DECnet, and Apple Computer’s AppleTalk.

Today, the most common network layer protocols are both from TCP/IP: IP version 4 (IPv4) and IP version 6 (IPv6).

The original host has a place to insert a value (a hexadecimal number) to identify the type of packet encapsulated inside the Ethernet frame.

However, what number should the sender put in the header to identify an IPv4 packet as the type? Or an IPv6 packet?

As it turns out, the IEEE manages a list of EtherType values, so that every network layer protocol that needs a unique EtherType value can have a number.

The sender just has to know the list. (To view the complete list; just go to www.ieee.org and search for EtherType or click here.)

For example, a host can send one Ethernet frame with an IPv4 packet and the next Ethernet frame with an IPv6 packet.

Each frame would have a different Ethernet Type field value, using the values reserved by the IEEE, as shown in the diagram below.

Ethernet also defines a way for nodes to find out whether a frame’s bits changed while crossing over an Ethernet link. (Usually, the bits could change because of some kind of electrical interference, or a bad NIC.)

Ethernet, like most data-link protocols, uses a field in the data-link trailer for the purpose of error detection.

The Ethernet Frame Check Sequence (FCS) field in the Ethernet trailer—the only field in the Ethernet trailer—gives the receiving node a way to compare results with the sender, to discover whether errors occurred in the frame.

The sender applies a complex math formula to the frame before sending it, storing the result of the formula in the FCS field.

The receiver applies the same math formula to the received frame. The receiver then compares its own results with the sender’s results. If the results are the same, the frame did not change; otherwise, an error occurred and the receiver discards the frame.

Note that error detection does not also mean error recovery.

Ethernet defines that the errored frame should be discarded, but Ethernet does not attempt to recover the lost frame.

Other protocols, notably TCP, recover the lost data by noticing that it is lost and sending the data again.

Ethernet LANs behave slightly differently depending on whether the LAN has mostly modern devices, in particular, LAN switches instead of some older LAN devices called LAN hubs.

Basically, the use of more modern switches allows the use of full-duplex logic, which is much faster and simpler than half-duplex logic, which is required when using hubs. 

Modern Ethernet LANs use a variety of Ethernet physical standards, but with standard Ethernet frames that can flow over any of these types of physical links.

Each individual link can run at a different speed, but each link allows the attached nodes to send the bits in the frame to the next node. They must work together to deliver the data from the sending Ethernet node to the destination node.

The process is relatively simple, on purpose; the simplicity lets each device send a large number of frames per second. The diagram below shows an example in which PC1 sends an Ethernet frame to PC2.

The steps in the diagram corresponds to below.

1. PC1 builds and sends the original Ethernet frame, using its own MAC address as the source address and PC2’s MAC address as the destination address.
2. Switch SW1 receives and forwards the Ethernet frame out its G0/1 interface (short for Gigabit interface 0/1) to SW2.
3. Switch SW2 receives and forwards the Ethernet frame out its F0/2 interface (short for Fast Ethernet interface 0/2) to PC2.
4. PC2 receives the frame, recognizes the destination MAC address as its own, and processes the frame.

The Ethernet network in the diagram uses full duplex on each link, but the concept might be difficult to see.

Full-duplex means that that the NIC or switch port has no half-duplex restrictions. So, to understand full duplex, you need to understand half duplex, as follows:

• Half duplex: The device must wait to send if it is currently receiving a frame; in other words, it cannot send and receive at the same time.
• Full duplex: The device does not have to wait before sending; it can send and receive at the same time.

So, with all PCs and LAN switches, and no LAN hubs, all the nodes can use full duplex.

All nodes can send and receive on their port at the same instant in time.

For example, in the above diagram, PC1 and PC2 could send frames to each other simultaneously, in both directions, without any half-duplex restrictions.

To understand the need for half-duplex logic in some cases, you have to understand a little about an older type of networking device called a LAN hub.

When the IEEE first introduced 10BASE-T in 1990, the Ethernet did not yet include LAN switches.

Instead of switches, vendors created LAN hubs. The LAN hub provided a number of RJ-45 ports as a place to connect links to PCs, just like a LAN switch, but it used different rules for forwarding data.

LAN hubs forward data using physical layer standards, and are therefore considered to be Layer 1 devices.

When an electrical signal comes in one hub port, the hub repeats that electrical signal out all other ports (except the incoming port).

By doing so, the data reaches all the rest of the nodes connected to the hub, so the data hopefully reaches the correct destination.

The hub has no concept of Ethernet frames, of addresses, and so on.

The downside of using LAN hubs is that if two or more devices transmitted a signal at the same instant, the electrical signal collides and becomes garbled.

The hub repeats all received electrical signals, even if it receives multiple signals at the same time.

For example, the diagram below shows the idea, with PCs Archie and Bob sending an electrical signal at the same instant of time (at Steps 1A and 1B) and the hub repeating both electrical signals out toward Larry on the left (Step 2).

If you replace the hub in diagram above with a LAN switch, the switch prevents the collision on the left.

The switch operates as a Layer 2 device, meaning that it looks at the data-link header and trailer.

A switch would look at the MAC addresses, and even if the switch needed to forward both frames to Larry on the left, the switch would send one frame and queue the other frame until the first frame was finished.

Now back to the issue created by the hub’s logic: collisions.

To prevent these collisions, the Ethernet nodes must use half-duplex logic instead of full-duplex logic.

A problem occurs only when two or more devices send at the same time; half-duplex logic tells the nodes that if someone else is sending, wait before sending.

In the diagram above, imagine that Archie began sending his frame early enough so that Bob received the first bits of that frame before Bob tried to send his own frame. Bob, at Step 1B, would notice that he was receiving a frame from someone else, and using half-duplex logic, would simply wait to send the frame listed at Step 1B.

Nodes that use half-duplex logic actually use a relatively well-known algorithm called carrier sense multiple access with collision detection (CSMA/CD).

The algorithm takes care of the obvious cases but also the cases caused by unfortunate timing.

For example, two nodes could check for an incoming frame at the exact same instant, both realize that no other node is sending, and both send their frames at the exact same instant, causing a collision.

CSMA/CD covers these cases as well, as follows:

• Step 1. A device with a frame to send listens until the Ethernet is not busy.
• Step 2. When the Ethernet is not busy, the sender begins sending the frame.
• Step 3. The sender listens while sending to discover whether a collision occurs; collisions might be caused by many reasons, including unfortunate timing. If a collision occurs, all currently sending nodes do the following:
  • A. They send a jamming signal that tells all nodes that a collision happened.
  • B. They independently choose a random time to wait before trying again, to avoid unfortunate timing.
  • C. The next attempt starts again at Step 1.

Although most modern LANs do not often use hubs, and therefore do not need to use half duplex, enough old hubs still exist in enterprise networks so that you need to be ready to understand duplex issues.

Each NIC and switch port has a duplex setting. For all links between PCs and switches, or between switches, use full duplex.

However, for any link connected to a LAN hub, the connected LAN switch and NIC port should use half-duplex. Note that the hub itself does not use half-duplex logic, instead just repeating incoming signals out every other port.

The diagram below shows an example, with full-duplex links on the left and a single LAN hub on the right. The hub then requires SW2’s F0/2 interface to use half-duplex logic, along with the PCs connected to the hub.