Ethernet Performance

Ethernet networks tend to be susceptible to performance problems as they grow, based on the CSMA/CD method of media access that they use. While implementing Layer 2 switching goes a long way towards better Ethernet performance, there are still a number of issues to consider when an Ethernet network begins to experience performance problems. Examples of reasons for congestion on Ethernet networks include having too many hosts on a given segment, not enough bandwidth available, broadcast storms, along with excessive broadcast or multicast traffic.

A few key metrics are used by Cisco to help decide when an Ethernet network is not performing at an appropriate level. These include:

  • Network utilization. A network utilization of over 40% on shared Ethernet segments represents that the network is saturated.

  • Broadcasts/multicasts. No Ethernet segment should have more than 20% combined broadcast and multicast traffic.

  • CRC Errors. There should be less than 1 CRC error per MB of network traffic.

  • Collisions. Less than 0.1% of network packets should be involved in collisions

While this data may seem difficult to obtain, a variety of network management tools can provide these particular metrics, along with many others. Examples include CiscoWorks and Cisco Netsys Performance Service Manager.

Ethernet Ports on a Cisco Router

A Cisco 2501 includes a single 10Mb Ethernet port. While many Cisco router models now include an integrated 10/100 RJ-45 port, the 2500 series uses what is referred to as a generic attachment unit interface (AUI) DB-15 port instead. The name for this connector (DB-15) comes from the fact that it is physically shaped like the letter ‘D’ and uses a 15-pin connector.

The purpose of providing an AUI port instead of a fixed RJ-45 port is flexibility – AUI ports use an external transceiver, which allow different media types to be connected, according to your network needs. Different transceivers can be attached that allow twisted pair, coaxial, or fiber connections. In that way, a transceiver connects to the DB-15 port, and then provides a port to which an RJ-45 or BNC connection could be made, for example.
But what is a transceiver? Well, its name comes from what it does – transmitting and receiving data. In most NICs that you’ll come across today, the transceiver is built directly into the network card. With older network cards (like those found on Ethernet 10Base5 networks) the transceiver was usually an external device.

When connecting a Cisco 2501’s Ethernet connection to a common RJ-45-based network, a transceiver attaches to the AUI port, and then a patch cable connects the transceiver to a switch or hub. Remember, a router connecting to a switch or hub always uses a straight-through cable, as shown in the figure below.

Figure: Connection between a Cisco 2501 and a switch or hub via an Ethernet transceiver. 

Physical and Data Link Layers – Frame Encapsulation

NetWare protocols can run over a variety of different network technologies, similar to TCP/IP. For the purpose of keeping things simple, in this section we’ll concentrate on Ethernet. However, IPX can also be run over Token Ring, FDDI, ATM and a variety of WAN technologies. If you recall, Novell originally defined the 802.3 frame type to be used over Ethernet networks. These frames lacked a Type field, since all 802.3 frames were assumed to be destined for IPX at the Network layer. The 802.3 frame type is commonly referred to as Novell Ethernet 802.3 or Ethernet Raw. Different versions of NetWare use different Ethernet frame types as their default for encapsulating IPX packets. In NetWare 3.11 and earlier versions, the default frame type was 802.3. Beginning with Netware 3.12, the default frame type was changed to 802.2.

This presents a technical challenge, since different frame types are not interoperable. As such, a system configured to use only 802.3 frames cannot communicate with a system in the same broadcast domain that uses only 802.2 frames.

To help overcome these limitations, a Cisco router can be configured to use more than one Ethernet encapsulation type for IPX on a single interface. This allows a router to communicate with systems using different frame types. We’ll look at the actual configuration of IPX on a router in Chapter 8. For now, it is sufficient to know the different frame types support by Cisco, and the terms used to describe them.

  • novell-ether. Refers to 802.3 frames that lack an LLC header.
  • sap. Refers to the IEEE standard Ethernet frame with an 802.2 LLC header.
  • arpa. Refers to the Ethernet II frame type.
  • snap. Refers to the Ethernet SNAP frame type.

Ethernet Physical Standards

Up to this point we’ve mainly been looking at the Data Link layer elements of Ethernet. However, Ethernet standards also define Physical layer characteristics such as cable distances, media types, and just about anything to do with physical connectivity, including connectors. You may already be familiar with some of the different varieties of Ethernet. They’re usually represented in a format such as 10BaseT or similar. Understanding the designations is critical, so we’ll look at these first.

When you see Ethernet defined as 10BaseT, you’re actually being provided with 3 pieces of information. In this example:

  • “10” defines the maximum speed of transmission in Megabits per second.
  • “Base” specifies that baseband transmission is used. Baseband transmission provides a single channel for digital transmission. In contrast, broadband transmission is analog and separates the cable into different frequency ranges or channels.
  • “T” defines that this type of Ethernet runs over twisted pair wiring. On a 10BaseT network, the minimum cable standard is Category 3.

A variety of different Ethernet standards exist, a cross section of which are outlined below.

  • 10Base2. 10Mbps Ethernet that runs over ThinNet coaxial cable. Maximum segment length of 185 meters and a maximum of 30 connected systems per segment.
  • 10Base5. 10Mbps Ethernet that runs over ThickNet coaxial cable. Maximum segment length of 500 meters and a maximum of 100 connected nodes per segment.
  • 10BaseF. 10Mbps Ethernet that runs over fiber optic cabling for distances up to 2 kilometers in full duplex.
  • 100BaseTX. Fast Ethernet (100Mbps) that runs over Cat5 twisted pair wiring. Maximum cable length is 100 meters.
  • 100BaseFX. Fast Ethernet that runs over fiber optic cabling.
  • 1000BaseT. Gigabit Ethernet (1000Mbps) that runs over Cat5 twisted pair wiring. Maximum cable length is 100 meters.
  • 1000BaseLX. Long wave Gigabit Ethernet over fiber. If using multimode fiber, maximum distance is 550 meters. If single mode fiber, maximum distance of approximately 5 kilometers.
  • 1000BaseSX. Short wave Gigabit Ethernet over fiber. Uses multimode fiber to span distances up to 550 meters.

Ethernet also makes use of features at the Physical layer by auto-negotiating elements such as link speed and duplex type when a network card is plugged into a switch or hub. Originally defined in the IEEE 802.3u specification (Fast Ethernet), this is accomplished using something called Fast Link Pulses (FLPs), which are sent between the system and the connected port. For example, you may have a network card that supports both 10 and 100 Mbps speeds. However, if the hub only supports 10 Mbps, they will negotiate the connection to the common setting (in this case 10 Mbps). The same is true for negotiation of the duplex type used. When using half duplex, a system can be either sending or receiving data, but not both concurrently. In full duplex, systems can send at receive at the same time.

Note that when plugged into a hub, systems will always communicate using half duplex, since they share the media and only one system can communicate at any given time. When a system is plugged directly into a switch port, full duplex becomes possible. To that end, it is worth noting that when you connect a hub to a switch, all computers plugged into that hub will automatically use half-duplex, since they’ll again be part of the same collision domain.

Ethernet Frame Types

Four different frame types exist in the world of Ethernet, mainly a result of different implementations created for different purposes at different times. Ethernet frame types include Ethernet 802.3, Ethernet 802.2, Ethernet II, and Ethernet SNAP. When two systems need to communicate on an Ethernet network, they must be using a common frame format. The confusion as to when a given Ethernet frame type is used is generally a result of different vendors and standards bodies moving in different directions. So how do you know which frame type will be used on a given network? That depends. Sometimes you can configure the interface to use a certain frame type with a certain upper-layer protocol (such as IPX). In other cases, a vendor or organization will choose the frame type to be used with a particular upper level protocol. For example, TCP/IP will always attempt to use Ethernet II, as defined by the Internet Engineering Task Force (IETF). For their Cisco Discovery Protocol (CDP), Cisco uses Ethernet SNAP framing. For the most part, network equipment will be able to handle multiple frame types on a single interface. In all cases, the minimum Ethernet frame size is 64 bytes, while the maximum size is 1518 bytes. Anything smaller than 64 bytes is invalid and referred to as a “runt”, while frames over 1518 bytes are also invalid and considered “giants”.

What’s the difference between the various frame types? Different frame types may include header fields that were created to address different technical challenges. In all Ethernet frame types you’ll find five main elements – a preamble, start of frame delimiter, header, data, and trailer. Each of these is described below:

  • Preamble. The purpose of the 7-byte preamble is to mark the beginning of a frame and to enable synchronization between a sender and receiver.
  • Start of Frame Delimiter. The 1-byte SOF field always ends in binary 11 to notify that the next bits represent the beginning of the destination MAC address.
  • Header. At a minimum, the header will contain the source and destination MAC addresses (6 bytes each), as well as an extra 2-byte field. Various frame types use this extra field differently, as we’ll discuss shortly.
  • Data. The data portion houses everything that was encapsulated by the upper-layer protocols prior to being passed down for framing.
  • Trailer. An Ethernet trailer consists of a Frame Check Sequence (FCS). This is where the Cyclic Redundancy Check (CRC) value is held that will be used to confirm that the frame has not been corrupted when it reaches its destination.

Now that we know the elements that are common to every frame, you need to be able to recognize the differences between the four main types.

Ethernet II

The Ethernet II frame type is by far the most simple. Those extra 2 bytes in the Ethernet header described earlier are used for a Type field in Ethernet II. The Type field simply identifies the upper layer protocol to which data should be passed. For example, a Type field of hex 0800 represents IP, while 8137 means that data is meant for IPX.

Figure: Ethernet II frame.

Ethernet 802.3

The Ethernet 802.3 frame was originally created by Novell for use with the IPX protocol, and was later standardized by the IEEE. Because these frames don’t contain any LLC information, they are sometimes referred to as Ethernet RAW. These frames contain a 2-byte Length field instead of a Type field – they automatically assume that the upper-layer protocol is IPX, and do not work with other upper layer protocols.

Figure: Ethernet 802.3 frame.

Ethernet 802.2 (SAP)

In order to provide a greater deal of flexibility with Ethernet framing, the IEEE defined what is known as the 802.2 Logical Link Control (LLC), the upper sub-layer of the Data Link Layer. At first glance an 802.2 frame may look like an 802.3 frame, since it has a length field. However, the first part of the data portion of an 802.2 Ethernet frame actually contains LLC information in the form of Source Service Access Point (SSAP), Destination Service Access Point (DSAP), and Control information.

Figure: Ethernet 802.2 frame.

As you’ll recall, the Logical Link Control SSAP and DSAP fields are used by the LLC sub-layer to interact with Network layer protocols. Examples of SAP codes include F0 for NetBIOS, 06 for IP, and E0 for IPX – again, all codes are represented in hexadecimal.

Ethernet SNAP

The final Ethernet frame type, Ethernet SNAP (which stands for Sub Network Access Protocol) was developed as a result of compatibility issues. Given that many vendors had been using the Ethernet II frame types for their upper layer protocols before 802.2 was standardized, they were left with a 1-byte SAP field where they had previously used a 2-byte Type field. This made moving to the new standard difficult, so the IEEE came up with the Ethernet SNAP frame type. Ethernet SNAP allows a higher degree of flexibility for proprietary protocols. The Ethernet SNAP frame type is commonly used with AppleTalk.

If you look at an Ethernet SNAP frame, you’ll notice that the SSAP and DSAP fields are always set to AA. These codes identify it as a SNAP frame. The command field always has a value of three, which specifies connectionless LLC service. Following this is the Organizationally Unique Identifier (OUI) field, which is used to define the organization that created the upper layer protocol. The Type field provides the same function as the Type field in an Ethernet II frame. When all is said and done, the SNAP information uses up 5 extra bytes of the data portion of an Ethernet frame.

Figure: Ethernet SNAP frame.

Ethernet Media Access and Addressing

Originally developed by Xerox in the 1970’s, Ethernet has become the defacto technology standard for LANs today. Digital, Intel, and Xerox (DIX) standardized Ethernet in 1980, with the IEEE version finalized in 1982 with the 802.3 standard. Over time Ethernet has undergone a number of changes, both with respect to how devices are connected, and the ways in which data is framed. A solid understanding of Ethernet concepts is imperative to your success on the CCNA and CCDA exams.

Media Access – CSMA/CD

The media access method used by Ethernet is the contention-based Carrier Sense Multiple Access with Collision Detection (CSMA/CD). This name not only defines the technology, but also describes how it works. “Carrier Sense” means that different devices are listening to the media for the opportunity to transmit. “Multiple Access” describes the media as being contention-based, in that it is shared amongst many computers. “Collision Detection” is an Ethernet feature whereby systems are capable of recognizing when a collision has occurred.

When a system uses collision detection techniques, there must be a way to try and avoid the same collision from happening repeatedly. CSMA/CD handles this by having systems back off for a random period of time after a collision occurs. If these collisions continue the system back off time will increase, considerably decreasing performance. Retransmission will be attempted up to 16 times before an error message will be passed to the upper-layer protocol in use. You may be familiar with some of the distance limitations imposed on Ethernet networks (we’ll look at those shortly). Understand, however, that the limitations exist not only because of signal attenuation, but also because as distances increase, the ability of CSMA/CD to properly detect collisions decreases. This is especially true when systems at opposite ends of a network attempt to communicate at the same time, sensing the media as available.

Note: Remember that CSMA/CD is the media access method used on Ethernet networks. As such, hosts on a traditional Ethernet network “share” the media, making their transmissions susceptible to collisions. Equipment like switches and bridges help to reduce network collisions, and will be looked at in more detail in Chapter 3.


All Ethernet network adapter cards are uniquely identified by a pre-assigned hardware (or MAC) address. A MAC address is a 48-bit address represented in hexadecimal format. The first 24 bits represent what is known as the Organizationally Unique Identifier (OUI), and represents a vendor code. The last 24 bits are assigned by the vendor and act as the unique identifier (and serial number) for a particular network card.

Tip: Remember that the first 24 bits of a MAC address identify a manufacturer like Cisco, and the last 24 bits represent the unique serial number of a card. For a complete list of manufacturer OUI codes, see

Hexadecimal is a numbering system that uses the numbers 0-9 and the letters A-F, where each hex digit represents 4 bits. As such, a MAC address will always be made up of 12 hex digits, in a format similar to 01-22-33-44-55-EF. The table below outlines the decimal value associates with each hexadecimal digit. Note that the highest valid hex digit is F – anything above that is not a valid character for a MAC address.

Hexadecimal to decimal conversions:


Tip: You can easily convert between decimal and hexadecimal by using the scientific calculator included with Microsoft Windows operating systems. A great online tool for converting numbers between decimal, hexadecimal, and binary easily can be found at

Note: During both the CCNA and CCDA exams, you will not have access to any type of calculator. As such, it is extremely important for you to be able to convert between decimal, hexadecimal, and binary numbers on your own.

Network Topologies

Network topologies describe both the physical and logical layouts of a network. A particular technology such as Token Ring might lead you to believe that the network is physically connected in one big circle of cable. In reality, the ring is formed in circuitry and the physical network appears as a star. Common topologies that you should be familiar with include Bus, Ring, Star, Hybrid, and Mesh.


A bus topology is by far the most simple. A bus is comprised of a single run of cable to which individual systems are attached. When a failure occurs on a bus network, it affects the entire network since the path for data transfer is disrupted. While you may still find Ethernet bus networks in smaller environments, they are becoming less and less common. In general, bus topologies are relatively inexpensive but tend to be more prone to failure.

Figure: Bus Network


As the name suggests, a ring topology is comprised of a number of systems connected in a type of loop. In most cases, systems connect to a hub-type device within which the actual ring is formed. For this reason, you’ll often see ring topologies described as being a star-ring, where data is passed through the network along the ring circuitry, but the physical layout actually appears to be a star. Logical ring / physical star layouts are commonly found in both Token Ring and FDDI environments.

Figure: Ring Network


By far the most common network topology found today, a physical star is created when systems connect to a central device such as a hub or switch. Systems branch out from this central device, creating the star-like appearance. The main benefit of this topology is the fact that a break in a cable only affects the particular connected system and not all others. Recognize, however, that a single point of failure still exists – if the hub or switch fails, all connected systems will not be able to communicate.

Figure: Star Network


Most large networks designed today tend to be variations of star topologies. However, many networks will be comprised of a number of different topologies rather than just one. For example, a company’s network might be a star-bus hybrid, where systems connect to hubs (forming the star) and then hubs interconnect using a bus. Various hybrids are possible, including stars, rings, and buses.

Figure: Hybrid Network 


Wide Area Networks (WANs) are often configured in a mesh topology for the purpose of redundancy. In a mesh, a router or switch may have more than one connection to a different site. In this way, if one link fails, another path exists. Because of this, mesh topologies are also much more expensive, and seldom found between computers on LANs (although certain systems requiring high-availability may be configured in this way). A mesh in which every system has a connection to every other system is referred to as a full mesh. If fewer connections exist, it is simply known as a partial mesh. A good example of a technology that is often configured in a mesh is Frame Relay.

Figure: Mesh Network 

Network Media Access Methods

There are a variety of methods by which data is merged onto a network, a concept referred to as the media access method. The media access method used depends on the way in which a particular technology such as Ethernet or Token Ring communicates. This section will look at the three most popular methods – contention-based, token passing, and polling.


Contention-based media access describes a way of getting data on to the network whereby systems ‘contend for’ or share the media. On a contention-based network, systems can only transmit when the media is free and clear of signals. In this way, devices listen to the media, and if no other system is transmitting, they can go ahead and send data. In cases where more than one system finds the network free and attempts to transmit, a data collision will occur, and systems will need to retransmit. On busy networks, the number of collisions can quickly get very high, adversely affecting performance. Remember that in this scenario, only a single system truly has access to the media at any given time, even though multiple systems may have data to send.

The best example of a contention-based network technology is Ethernet, which uses a scheme called Carrier Sense Multiple Access with Collision Detection (CSMA/CD). The fact that Ethernet is contention-based is a reason why many people thought that the technology would never be a good solution for large networks. As time passed, different techniques were developed to provide a way for contention-based networks to scale to larger sizes. A great example is the use of switches to segment a network, thus significantly reducing (or even eliminating) collisions.

Token Passing

A more orderly scheme for moving data between network systems is found when token passing is used. In token-passing media access environments, a special frame referred to as a token repeatedly circles the network, passed from system to system. If a system has control of the token, it can transmit data. If it doesn’t, it must wait for the token to become available again.

While this might sound like a very slow way to go about passing data, it’s important to understand that the token moves around the network at incredibly high speeds. Understand also that because this method isn’t contention based, there won’t be any collisions, further increasing performance

Examples of technologies that use token-passing media access include Token Ring and Fiber Distributed Data Interface (FDDI), both of which will be described in detail later in this chapter.


While contention and token-passing methods are by far the most popular ways in which PCs access LAN media, some technologies rely on a technique called polling. Polling is a deterministic way of allowing systems access to the network while also avoiding collisions. When used, a device referred to as the master polls systems to see if they have data to transmit. In this way, polling is similar to token passing, except that the central device controls the order in which systems are contacted. The downside of polling is that when the master device fails, the network fails. Most popular in mainframe and minicomputer environments, polling is a technique used in protocols such as Synchronous Data Link Control (SDLC).

Shielded Twisted Pair and Fiber Optic Cabling

Shielded Twisted Pair (STP) cabling is very similar to UTP, with one key difference. STP incorporates an additional conductive foil shielding around each pair of wires. While this helps further cut down on EMI, STP is more expensive and can also be thicker and harder to install than UTP cables. STP cabling is most often used on Token Ring networks.

Fiber optic cabling is another option for network media, though considerably more expensive. This is not only in terms of the cost of fiber optic cables, but also related components such as network cards and switching equipment. Unlike UTP and STP, which run over copper wires, fiber optic cabling instead sends pulses of light through a pure glass core encased within Kevlar sheathing. The fact that light isn’t susceptible to EMI provides great advantages, especially in places with a high degree of possible interference such as factories or warehouses. Fiber optic cabling also allows data to be transmitted along much greater distances that traditional copper wire, in some cases up to hundreds of kilometers. Using fiber does have some disadvantages outside just cost – it’s significantly more difficult to install, and subject to various bending limitations. It is worth noting that fiber cables will have two separately encased fiber stands, one used for transmitting and the other used for receiving. In other words, the strand connected to the transmit port on one device will be connected to the receive port on the other device.

Different types of connectors can be used to connect to fiber optic ports. The most common are round plug-style ST connectors and square block-style SC connectors. It’s important that you make sure that you’ve purchased the correct cables to match the ports on your particular equipment.

The two main types of fiber cabling used on networks are single mode and multimode. Single mode fiber can span much greater distances and carries a single ray of light up to a number of kilometers. There are a number of factors involved in the distances that fiber optic cabling can span, though it mainly relates the micron diameter of the glass core. To be clear, a micron is one millionth of a meter. The most common diameter for fiber optic cabling is 62.5 microns. As the diameter on the cable decreases, the distance that can be spanned increases. Single mode fiber is most commonly used on networks that employ long-wavelength optics (LX).

Multimode fiber can also be used in long-wavelength optics, but is more commonly used with short-wavelength (SX) technologies. Multimode fiber carries many different light signals at once, each at different angles of refraction. It can only travel shorter distances, up to approximately 550 meters.

Fiber optic cabling is becoming popular on LANs, especially for use with Gigabit Ethernet. While fiber optic connections to the desktop aren’t common, you’ll often see them used for backbone or trunk connections between switches, and frequently for connections to servers as well.

Wiring Straight and Crossover Ethernet Cables

Although you may think it unlikely, I can guarantee that you’ll find plenty of network support people in the world who don’t know how to wire a simple patch cable properly. Understanding the roles of the wires in twisted pair cabling goes a long way towards explaining the communication process between connected systems. What we’re going to look at here is the way in which different network cables are created using Cat 5.

Even though there are 8 wires (4 pairs) in Cat 5 wiring, only 2 of those pairs are used to transmit and receive data. You may have heard the term ‘tip and ring’, especially in the world of telephony. The four wires that transmit and receive data are considered to be the tip, while the other wires provide grounding and are referred as the ring. When using Cat 5 cabling, there are two possible transmitting and receiving channels, allowing systems to communicate in full duplex if plugged into a switch. Full-duplex means that systems can both send and receive data at the same time – note that many older network cards will only support half-duplex, while newer cards almost certainly will support both. Full- and half-duplex will be looked at in detail when we discuss Ethernet.

Like just about anything else in the networking world, there are standards for wiring UTP cables. These standards are defined by the Electronic Industries Alliance/Telecommunications Industry Association (EIA/TIA) and fall into two flavors – 568A and 568B. For the most part, the B standard is more popular. However, many government contracts require that the A standard be used. To be honest, it’s well worth knowing both, as will become clear in a moment. When creating network cables, you’ll notice that they are made up of individual wires of different colors. Some appear as a solid color, while others appear white with a colored stripe. The proper terminology is to call the orange wire ‘orange’ while calling the white wire with the orange stripe ‘white-orange’ – the background color should always be specified first.

A straight-through cable has two ends that are identical. When looking at cables, compare the RJ-45 ends side-by-side with the snap-in clip facing down. If you’re looking at a straight-through cable the ends will be the same, while on a crossover cable, the ends will be different. To be even more precise, a crossover cable simply has one end wired according to the A standard, and the other wired to the B standard – that’s why knowing both standards is so useful.
The tables below outline the wiring order for 568A and 568B. I’ve included both the pin numbers and the colors to help you along. Note that the pin numbers represent the individual wires from left to right, when looking at an RJ-45 connector with the clip facing down. The leftmost will always be pin 1, and the rightmost pin 8.

Note: Some vendors engage in the bad habit of not wiring their pre-packages cables to the 568A and 568B standard. It’s important to recognize that the wire numbering is most important during the communication process, not the colors. However, for the Cisco exams you should be familiar with the wire color and numbering standards listed here.

568A and 568B Wiring tables:

1. White-Green
2. Green
3. White-Orange
4. Blue
5. White-Blue
6. Orange
7. White Brown
8. Brown

1. White-Orange
2. Orange
3. White-Green
4. Blue
5. White-Blue
6. Green
7. White-Brown
8. Brown
If you look closely, you’ll notice that only the orange and green pairs switch places between the standards.

To understand what the pins do, let’s take a look at a straight-through cable. In the example below, note that the cable is plugged into a PC at one end and a hub at the other. Note that the pin-outs on the PC and hub have different roles, and that only 4 wires are actually used in the communication process – pins 1, 2, 3 and 6:

Figure: Straight-through connection, PC to hub.