Fast Ethernet

Fast Ethernet

When Ethernet technology availed itself to users, the 10 Mbps bandwidth seemed like an unlimited resource. (Almost like when we had 640k of PC RAM…it seemed we would never need more!) Yet workstations have developed rapidly since then, and applications demand more data in shorter amounts of time. When the data comes from remote sources rather than from a local storage device, this amounts to the application needing more network bandwidth. New applications find 10 Mbps to be too slow. Consider a surgeon downloading an image from a server over a 10 Mbps shared media network. He needs to wait for the image to download so that he can begin/continue the surgery. If the image is a high resolution image, not unusually on the order of 100 MB, it could take a while to receive the image. What if the shared network makes the available user bandwidth about 500 kbps (a generous number for most networks) on the average? It could take the physician 26 minutes to download the image:

100 MB x 8/500 kbps = 26 minutes

If that were you on the operating table waiting for the image to download, you would not be very happy! If you are the hospital administration, you are exposing yourself to surgical complications at worst and idle physician time at best. Obviously, this is not a good situation. Sadly, many hospital networks function like this and consider it normal. Clearly, more bandwidth is needed to support this application.

Recognizing the growing demand for higher speed networks, the IEEE formed the 802.3u committee to begin work on a 100 Mbps technology that works over twisted-pair cables. In June of 1995, IEEE approved the 802.3u specification defining a system that offered vendor interoperability at 100 Mbps.

Like 10 Mbps systems such as 10BaseT, the 100 Mbps systems use CSMA/CD, but provide a tenfold improvement over legacy 10 Mbps networks. Because they operate at 10 times the speed of 10 Mbps Ethernet, all timing factors reduce by a factor of 10. For example, the slotTime for 100 Mbps Ethernet is 5.12 microseconds rather than 51.2 microseconds. The IFG is .96 microseconds. And because timing is one tenth that of 10 Mbps Ethernet, the network diameter must also shrink to avoid late collisions.

An objective of the 100BaseX standard was to maintain a common frame format with legacy Ethernet. Therefore, 100BaseX uses the same frame sizes and formats as 10BaseX. Everything else scales by one tenth due to the higher data rate. When passing frames from a 10BaseX to a 100BaseX system, the interconnecting device does not need to re-create the frame's Layer 2 header because they are identical on the two systems.

10BaseT, the original Ethernet over twisted-pair cable standard, supports Category 3, 4, and 5 cables up to 100 meters in length. 10BaseT uses a single encoding technique, Manchester, and signals at 20 MHz well within the bandwidth capability of all three cable types. Because of the higher signaling rate of 100BaseT, creating a single method to work over all cable types was not likely. The encoding technologies that were available at the time forced IEEE to create variants of the standard to support Category 3 and 5 cables. A fiber optic version was created as well.

Full-Duplex and Half-Duplex Support

This chapter began with discussions on legacy Ethernet and CSMA/CD. Legacy Ethernet uses CSMA/CD because it operates on a shared media where only one device can talk at a time. When a station talks, all other devices must listen or else the system experiences a collision. In a 10 Mbps system, the total bandwidth available is dedicated to transmitting or receiving depending upon whether the station is the source or the recipient. This describes half duplex.

The original LAN standards operate in half-duplex mode allowing only one station to transmit at a time. This was a side effect of the bus topology of 10Base5 and 10Base2 where all stations attached to the same cable. With the introduction of 10BaseT, networks deployed hubs and attached stations to the hub on dedicated point-to-point links. Stations do not share the wire in this topology. 100BaseX uses hubs with dedicated point-to-point links. Because each link is not shared, a new operational mode becomes feasible. Rather than running in half-duplex mode, the systems can operate in full-duplex mode, which allows stations to transmit and receive at the same time, eliminating the need for collision detection. What advantage does this provide? The tremendous asset of the precious network commodity bandwidth. When a station operates in full-duplex mode, the station transmits and receives at full bandwidth in each direction.

The most bandwidth that a legacy Ethernet device can expect to enjoy is 10 Mbps. It either listens at 10 Mbps or transmits at 10 Mbps. In contrast, a 100BaseX device operating in full-duplex mode sees 200 Mbps of bandwidth 100 Mbps for transmitting and 100 Mbps for receiving. Users upgraded from 10BaseT to 100BaseX have the potential to immediately enjoy a twentyfold, or more, bandwidth improvement. If the user previously attached to a shared 10 Mbps system, they might only practically enjoy a couple megabits per second of effective bandwidth. Upgrading to a full duplex 100 Mbps system might provide a perceived one hundredfold improvement. If your users are unappreciative of the additional bandwidth, you have an unenviable group of colleagues with which to work. Put them back on 10BaseT!

Note

Be aware, however: Just because an interface card runs 100BaseX full duplex, you cannot assume that the device where you install it supports full-duplex mode. In fact, some devices might actually experience worse throughput when in full-duplex mode than when in half-duplex mode. For example, Windows NT 4.0 does not support full-duplex operations because of driver limitations. Some SUN workstations can also experience this, especially with Gigabit Ethernet.

The IEEE 802.3x committee designed a standard for full-duplex operations that covers 10BaseT, 100BaseX, and 1000BaseX. (1000BaseX is Gigabit Ethernet discussed in a later section, "Gigabit Ethernet.") 802.3x also defined a flow control mechanism. This allows a receiver to send a special frame back to the source whenever the receiver's buffers overflow. The receiver sends a special packet called a pause frame. In the frame, the receiver can request the source to stop sending for a specified period of time. If the receiver can handle incoming traffic again before the timer value in the pause frame expires, the receiver can send another pause frame with the timer set to zero. This tells the receiver that it can start sending again.

Although 100BaseX supports both full- and half-duplex modes, you can deploy 100 Mbps hubs that operate in half-duplex mode. That means the devices attached to the hub share the bandwidth just like the legacy Ethernet systems. In this case, the station must run in half-duplex mode. To run in full-duplex mode, the device and the hub (switch) must both support and be configured for full duplex. Note that you cannot have a full duplex for a shared hub. If the hub is shared, it must operate in half-duplex mode.

Autonegotiation

With the multiple combinations of network modes available, configuring devices gets confusing. You need to determine if the device needs to operate at 10 or 100 Mbps, whether it needs to run in half- or full-duplex mode, and what media type to use. The device configuration must match the hub configuration to which it attaches.

Autonegotiation attempts to simplify manual configuration requirements by enabling the device and hub to automatically agree upon the highest common operational level. The 802.3u committee defined Fast Link Pulse (FLP) to support the autonegotiation process. FLP, an enhanced version of 10BaseT's Link Integrity Test, sends a series of pulses on the link announcing its capabilities. The other end also transmits FLP announcements, and the two ends settle on whatever method has highest priority in common between them. Table 1-2 illustrates the priority scheme.

Table 1-2. Autonegotiation Prioritization

Priority

Method

1

100BaseT2 full duplex

2

100BaseT2 half duplex

3

100BaseTX full duplex

4

100BaseT4 (Only half duplex)

5

100BaseTX half duplex

6

10BaseT full duplex

7

10BaseT half duplex

According to Table 1-2, 100BaseT2 full-duplex mode has highest priority, whereas the slowest method, 10BaseT half-duplex, has lowest priority. Priority is determined by speed, cable types supported, and duplex mode. A system always prefers 100 Mbps over 10 Mbps, and always prefers full duplex over half duplex. Note that 100BaseT2 has higher priority than 100BaseTX. This is not a direct result of 100BaseT2 being a more recent medium. Rather, 100BaseT2 has higher priority because it supports more cable types than does 100BaseTX. 100BaseTX only supports Category 5 type cable, whereas 100BaseT2 supports Category 3, 4, and 5 cables.

Tip

Not all devices perform autonegotiation. We have observed at several customer locations failure of the autonegotiation process either because of equipment not supporting the feature or poor implementations. We recommend that critical devices such as routers, switches, bridges, and servers be manually configured at both ends of the link to ensure that, upon reboot, the equipment operates in a common mode with its hub/switch port.

100BaseTX

Many existing 10 Mbps twisted-pair systems use a cabling infrastructure based upon Category 5 (unshielded twisted-pair) UTP and (shielded twisted-pair) STP. The devices use two pairs of the cable: one pair on pins 1 and 2 for transmit and one pair on pins 3 and 6 for receive and collision detection. 100BaseTX also uses this infrastructure. Your existing Category 5 cabling for 10BaseT should support 100BaseTX, which also implies that 100BaseTX works up to 100 meters, the same as 10BaseT.

100BaseTX uses an encoding scheme like Fiber Distributed Data Interface (FDDI) of 4B/5B. This encoding scheme adds a fifth bit for every four bits of user data. That means there is a 25 percent overhead in the transmission to support the encoding. Although 100BaseTX carries 100 Mbps of user data, it actually operates at 125 Megabaud. (We try not to tell this to marketing folks so that they do not put on their data sheets 125 Mbps throughput!)

100BaseT4

Not all building infrastructures use Category 5 cable. Some use Category 3. Category 3 cable was installed in many locations to support voice transmission and is frequently referred to as voice grade cable. It is tested for voice and low speed data applications up to 16 MHz. Category 5 cable, on the other hand, is intended for data applications and is tested at 100 MHz. Because Category 3 cable exists in so many installations, and because many 10BaseT installations are on Category 3 cable, the IEEE 802.3u committee included this as an option. As with 10BaseT, 100BaseT4 links work up to 100 meters. To support the higher data rates, though, 100BaseT4 uses more cable pairs. Three pairs support transmission and one pair supports collision detection. Another technology aspect to support the high data rates over a lower bandwidth cable comes from the encoding technique used for 100BaseT4. 100BaseT4 uses an encoding method of 8B/6T (8 bits/6 ternary signals) which significantly lowers the signaling frequency, making it suitable for voice-grade wire.

100BaseT2

Although 100BaseT4 provides a solution for Category 3 cable, it needs four pairs to support operations. Most Category 3 cable installations intend for the cable to support voice communications. By consuming all the pairs in the cable for data transmissions, no pairs remain to support voice communications. 100BaseT2, completed by IEEE in 1997 and called 802.3y, operates on Category 3, 4, and 5 cables and only requires two cable pairs. A new addition to the 100BaseT standards, 100BaseT2 relies upon advanced digital signal processing chips and encoding methods called PAM 5x5 (4 bits point to one of 25 possible values) to function over the lower bandwidth cable type. 100BaseT2 works on link lengths up to 100 meters.

100BaseFX

802.3u specifies a variant for single-mode and multimode fiber optic cables. 100BaseFX uses two strands (one pair) of fiber optic cables one for transmitting and one for receiving. Like 100BaseTx, 100BaseFX uses a 4B/5B encoding signaling at 125 MHz on the optical fiber. When should you use the fiber optic version? One clear situation arises when you need to support distances greater than 100 meters. Multimode supports up to 2,000 meters in full-duplex mode, 412 meters in half-duplex mode. Single-mode works up to 10 kms a significant distance advantage. Other advantages of fiber include its electrical isolation properties. For example, if you need to install the cable in areas where there are high levels of radiated electrical noise (near high voltage power lines or transformers), fiber optic cable is best. The cable's immunity to electrical noise makes it ideal for this environment. If you are installing the system in an environment where lightning frequently damages equipment, or where you suffer from ground loops between buildings on a campus, use fiber. Fiber optic cable carries no electrical signals to damage your equipment.

Table 1-3. 100BaseX Media Comparisons

Standard

Cable Type

Mode

Pairs Required

Distance (meters)

10BaseT

Category 3,4,5

Half Duplex

2

100

100BaseTX

Category 5

Half Duplex, Full Duplex

2

100

100BaseT4

Category 3

Half Duplex

4

100

100BaseT2

Category 3,4,5

Half Duplex, Full Duplex

2

100

100BaseFX

Multimode

Half Duplex, Full Duplex

1

412(Half-Duplex) 2000(Full-Duplex)

100BaseFX

Single-mode

Half Duplex, Full Duplex

1

10 kms

Note that the multimode fiber form of 100BaseFX specifies two distances. If you run the equipment in half-duplex mode, you can only transmit 412 meters. Full-duplex mode reaches up to 2 kms.

Media-Independent Interface (MII)

When you order networking equipment, you usually order the system with a specific interface type. For example, you can purchase a router with a 100BaseTX connection. When you buy it with this kind of interface, the 100BaseTX transceiver is built in to the unit. This connection is fine, as long as you only attach it to another 100BaseTX device such as another workstation, hub, or switch. What if you decide at a later time that you need to move the router to another location, but the distance demands that you need to connect over fiber optics rather than over copper? You need to buy another module to replace the 100BaseTX that you previously installed. This can be costly.

An alternative is the MII connector. This is a 40-pin connector that allows you to connect an external transceiver that has an MII connection on one side and a 100BaseX interface on the other side. Functionally, it is similar to the AUI connector for 10 Mbps Ethernet and allows you to change the media type without having to replace any modules. Rather, you can change a less expensive media adapter (transceiver). For Fast Ethernet, if you decide to change the interface type, all you need to do is change the MII transceiver. This is potentially a much less expensive option than replacing an entire router module.

Network Diameter (Designing with Repeaters in a 100BaseX Network)

In a legacy Ethernet system, repeaters extend cable distances, allowing networks to reach further than the segment length. For example, a 10Base2 segment only reaches 185 meters in length. If an administrator desires to attach devices beyond this reach, the administrator can use repeaters to connect a second section of 10Base2 cable to the first. In a 10BaseT network, hubs perform the repeater functions allowing two 100 meter segments to connect together. Legacy repeaters are discussed in more detail in Chapter 2, "Segmenting LANs."

802.3u defines two classes of repeaters for 100BaseX systems. The two repeater classes differ in their latency which affects the network diameter supported. A Class I repeater latency is 0.7 microseconds or less, whereas a Class II repeater latency is 0.46 microseconds or less. Only one Class I repeater is allowed in a 100BaseX system, whereas two hops are permitted for Class II repeaters. Why are there two repeater classes?

Class I repeaters operate by converting the incoming signal from a port into an internal digital signal. It then converts the frame back into an analog signal when it sends it out the other ports. This allows a Class I repeater to have a mix of ports that are 100BaseTX, 100BaseT4, 100BaseT2 or 100BaseFX. Remember that the line encoding scheme for these methods differ. The only ones with a common encoding scheme, 4B/5B, are 100BaseTX and 100BaseFX. A Class I repeater can translate the line encoding to support the differing media types.

Class II repeaters, on the other hand, are not as sophisticated. They can only support ports with a same line encoding method. Therefore, if you are using 100BaseT4 cabling, all ports in a Class II repeater must be 100BaseT4. Similarly, if you are using 100BaseT2, all ports of your Class II repeater must be 100BaseT2. The only exception to mixing is for 100BaseTX and 100BaseFX, because these both use 4B/5B and no encoding translation is necessary.

The lower latency value for a Class II repeater enables it to support a slightly larger network diameter than a Class I based network. Converting the signal from analog to digital and performing line encoding translation consumes bit times. A Class I repeater therefore introduces more latency than a Class II repeater reducing the network diameter.

Figure 1-4 illustrates interconnecting stations directly together without the use of a repeater. Each station is referred to as a DTE (data terminal equipment) device. Transceivers and hubs are DCE (data communication equipment) devices. Use a straight through cable when connecting a DTE to a DCE device. Use a cross-over when connecting a DTE to DTE or a DCE to DCE. Either copper or fiber can be used. Be sure, however, that you use a cross-over cable in this configuration. A cross-over cable attaches the transmitter pins at one end to the receiver pins at the other end. If you use a straight through cable, you connect "transmit" at one end to "transmit" at the other end and fail to communicate. (The Link Status light does not illuminate!)

Figure 1-4. Interconnecting DTE to DTE

Note

There is an exception to this where you can, in fact, connect two DTE or two DCE devices directly together with a straight through cable. Some devices have MDI (medial interface) and MDIX ports. The MDIX is a media interface cross-over port. Most ports on devices are MDI. You can use a straight through cable when connecting from an MDI to an MDIX port.

Using a Class I repeater as in Figure 1-5 enables you to extend the distance between workstations. Note that with a Class I repeater you can mix the types of media attaching to the repeater. Any mix of 100BaseTX, 100BaseT4, 100BaseT2, or 100BaseFX works. Only one Class I repeater is allowed in the network. To connect Class I repeaters together, a bridge, switch, or router must connect between them.

Figure 1-5. Networking with Class I Repeaters

Class II repeaters demand homogenous cabling to be attached to them. If you use 100BaseT4, all ports must be 100BaseT4. The only mix permitted uses 100BaseTX and 100BaseFX. Figure 1-6 illustrates a network with only one Class II repeater.

Figure 1-6. Networking with one Class II Repeater

Unlike Class I repeaters, two Class II repeaters are permitted as in Figure 1-7. The connection between the repeaters must be less than or equal to five meters. Why daisy chain the repeaters if it only gains five meters of distance? Simply because it increases the number of ports available in the system.

Figure 1-7. Networking with Two Class II Repeaters

The networks in Figure 1-5 through Figure 1-7 illustrate networks with repeaters operating in half-duplex mode. The network diameter constraints arise from a need to honor the slotTime window for 100BaseX half-duplex networks. Extending the network beyond this diameter without using bridges, switches, or routers violates the maximum extent of the network and makes the network susceptible to late collisions. This is a bad situation. The network in Figure 1-8 demonstrates a proper use of Catalyst switches to extend a network.

Figure 1-8. An Extended 100BaseX Network with Catalyst Switches

Practical Considerations

100BaseX networks offer at least a tenfold increase in network bandwidth over shared legacy Ethernet systems. In a full-duplex network, the bandwidth increases by twentyfold. Is all this bandwidth really needed? After all, many desktop systems cannot generate anywhere near 100 Mbps of traffic. Most network systems are best served by a hybrid of network technologies. Some users are content on a shared 10 Mbps system. These users normally do little more than e-mail, Telnet, and simple Web browsing. The interactive applications they use demand little network bandwidth and so the user rarely notices delays in usage. Of the applications mentioned for this user, Web browsing is most susceptible because many pages incorporate graphic images that can take some time to download if the available network bandwidth is low.

If the user does experience delays that affect work performance (as opposed to non-work-related activities), you can increase the users bandwidth by doing the following:

• Upgrading the user to 10BaseT full duplex and immediately double the bandwidth.

• Upgrading the user to 100BaseX half duplex.

• Upgrading the user to 100BaseX full duplex.

Which of these is most reasonable? It depends upon the user's application needs and the workstation capability. If the user's applications are mostly interactive in nature, either of the first two options can suffice to create bandwidth.

However, if the user transfers large files, as in the case of a physician retrieving medical images, or if the user frequently needs to access a file server, 100BaseX full duplex might be most appropriate. Option 3 should normally be reserved for specific user needs, file servers, and routers.

Another appropriate use of Fast Ethernet is for backbone segments. A corporate network often has an invisible hierarchy where distribution networks to the users are lower speed systems, whereas the networks interconnecting the distribution systems operate at higher rates. This is where Fast Ethernet might fit in well as part of the infrastructure. The decision to deploy Fast Ethernet as part of the infrastructure is driven by corporate network needs as opposed to individual user needs, as previously considered. Chapter 8, "Trunking Technologies and Applications," considers the use of Fast Ethernet to interconnect Catalyst switches together as a backbone.