In this section, you’ll learn the difference between the Ethernet media types and how to use them in your internetworks. We’ll cover the following Ethernet types:
10BaseT stands for 10 million bits per second (Mbps), baseband technology, twisted-pair. This Ethernet technology has the highest install base of any network in the world. It runs the Carrier Sense Multiple Access/Collision Detection (CSMA/CD) protocol and, if correctly installed, is an efficient network. However, if it gets too large and the network is not segmented correctly, problems occur. It is important to understand collision and broadcast domains and how to correctly design the network with switches and routers.
10BaseT Ethernet is typically used only at the access layer, and even then, FastEthernet (100BaseT) is quickly replacing it as the prices for 100BaseT continue to drop. It would be poor design to place 10BaseT at the distribution or core layers. You need transits that are much faster than 10BaseT at these layers.
The distance that 10BaseT can run and be within specification is 100 meters (330 feet). The 100 meters includes the following:
Five meters from the switch to the patch panel
Ninety meters from the patch panel to the office punch-down block
Five meters from the punch-down block to the desktop connection
This doesn’t mean that you can’t run more than 100 meters on a cable run; it just is not guaranteed to work.
FastEthernet is 10 times faster than 10Mbps Ethernet. The great thing about FastEthernet is that, like 10BaseT, it is still based on the CSMA/CD signaling. This means that you can run 10BaseT and 100BaseT on the same network without any problems. What a nice upgrade path this type of network can give you. You can put all your clients on 10BaseT and upgrade only the servers to 100BaseT if you need to. However, you can’t even buy a PC that doesn’t have a 10/100 Ethernet card in it anymore, so you really don’t need to worry about compatibility and speed issues from the user’s perspective.
FastEthernet works great at all layers of the hierarchical model. It can be used to give high performance to PCs and other hosts at the access layer, provide connectivity from the access layer to the distribution layer switches, and connect the distribution layer switches to the core network. Connecting a server block to the core layer would need, at a minimum, FastEthernet or maybe even Gigabit Ethernet.
There are two different specifications for FastEthernet, but the IEEE 802.3u is the most popular. The 802.3u specification is 100Mbps over category 3 or 5, twisted-pair (typically just category 5 or 5-plus is used for FastEthernet). The second Ethernet specification, called 802.12, used a different signaling technique, called Demand Priority Access Method (DPAM), which was more efficient than the CSMA/CD access method. The IEEE passed both methods in June 1995, but because 802.3 Ethernet had such a strong name in the industry, 802.12—also called 100VG- AnyLAN—has virtually disappeared from the market. As with the Macintosh and NetWare operating systems, it doesn’t mean anything if you have a better product; it matters only how you market it.
The IEEE 802.3u committee’s goals can be summarized as follows:
Provide seamless integration with the installed base
Provide 100BaseT at only two times (or less) the cost of 10BaseT
Increase aggregate bandwidth
Provide multiple-vendor standardization and operability
Provide time-bounded delivery
Precisely speaking, 802.12 is usually referred to as 100VG-AnyLAN. 100 is for 100Mbps, VG is for voice-grade cable, and AnyLAN is because it was supposed to be able to use either Ethernet or token-ring frame formats. The main selling point—the use of all four pairs of voice-grade cable—was also its main drawback. This feature is useful if all you have is VG, but it’s overshadowed completely by 100BaseT if you have category 5 cable or better. Developed at the time that category 5 was becoming popular, wide-scale implementations of new cabling systems just completely sidelined 802.12.
FastEthernet requires a different interface than 10BaseT Ethernet. 10Mbps Ethernet used the Attachment Unit Interface (AUI) to connect Ethernet segments. This provided a decoupling of the MAC layer from the different requirements of the various Physical layer topologies, which allowed the MAC to remain constant but meant the Physical layer could support any existing and new technologies. However, the AUI interface could not support 100Mbps Ethernet because of the high frequencies involved. 100BaseT needed a new interface, and the Media Independent Interface (MII) provides it.
100BaseT actually created a new subinterface between the Physical layer and the Data Link layer, called the Reconciliation Sublayer (RS). The RS maps the 1s and 0s to the MII interface. The MII uses a nibble, which is defined as 4 bits. AUI used only 1 bit at a time. Data transfers across the MII at one nibble per clock cycle, which is 25MHz. 10Mbps uses a 2.5MHz clock.
Full-duplex Ethernet can both transmit and receive simultaneously and uses point-to-point connections. It is typically referred to as “collision free” because it doesn’t share bandwidth with any other devices. Frames sent by two nodes can not collide because there are physically separate transmit and receive circuits between the nodes.
Both 10Mbps and 100Mbps Ethernet use four of the eight pins available in standard category 5 UTP cable. Pin 1 on one side and pin 3 on the other are linked, as are pins 2 and 6. When the connection is configured for half-duplex, the data can flow in only one direction at a time, while with full-duplex, data can come and go without collisions because the receive and send channels are separate.
Full-duplex is available when connected to a switch but not to a hub. Full-duplex is also available on 10Mbps, 100Mbps, and Gigabit Ethernet. Because it eliminates collisions, a full- duplex connection will disable the collision detection function on the port.
Full-duplex Ethernet provides equal bandwidth in both directions. But because users typically work with client/server applications using read/write asymmetrical traffic, arguably the best performance increase gained by full-duplex connectivity would be in the distribution layer, not necessarily in the access layer. Nonetheless, the ease with which it can be implemented and the increase in throughput—no matter how incremental—means that many networks run full- duplex throughout the network.
Full-duplex with flow control was created to avoid packets being dropped if the buffers on an interface fill up before all packets can be processed. However, some vendors might not interoperate, and the buffering might have to be handled by upper-layer protocols instead.
Auto-negotiation is a process that enables clients and switches to agree on a link capability. This is used to determine the link speed as well as the duplex being used. The auto-negotiation process uses priorities to set the link configuration. Obviously, if both a client and switch port can use 100Mbps, full-duplex connectivity, that would be the highest-priority ranking, whereas half-duplex, 10Mbps Ethernet would be the lowest ranking.
Auto-negotiation uses Fast Link Pulse (FLP), which is an extension to the Normal Link Pulse (NLP) standard used to verify link integrity. NLP is part of the original 10BaseT standard. Commonly, these auto-negotiation protocols do not work that well and you would be better off to configure the switch and NICs to run in a dedicated mode instead of letting the clients and switches auto-negotiate. Later in this chapter, we’ll show you how to configure your switches with both the speed and duplex options.
Auto-negotiation is one of the most common causes of frame check sequence (FCS) and alignment errors. If two devices are connected, and one is set to full-duplex and the other to half-duplex, one is sending and receiving on the same two wires while the other is using two wires to send and two to receive. Statically configuring the duplex on the ports eliminates this problem.
Intermittent connectivity issues can often be traced to auto-negotiation problems. If a single user occasionally has long connectivity outages, statically setting speed and duplex on both ends often helps.
FastEthernet does have some drawbacks. It uses the same signaling techniques as 10Mbps Ethernet, so it has the same distance constraints. In addition, 10Mbps Ethernet can use up to four repeaters, whereas FastEthernet can use only one or two, depending on the type of repeater. Table 2.1 shows a comparison of FastEthernet technologies.
Of course, the issue of the number of Ethernet repeaters in use is really only of concern when using a hub-based half-duplex system. Once we move to a switched Ethernet environment, the collision domains are considerably reduced in size and we don’t need repeaters, and the use of full-duplex Ethernet removes the need to detect collisions entirely, changing the CSMA/ CD operation to just CSMA.
In the corporate market, Gigabit Ethernet is the new hot thing. What is so great about Gigabit is that it can use the same network that your 10Mbps and 100Mbps Ethernet now use. You certainly do have to worry about distance constraints, but what a difference it can make in just a server farm alone!
Just think how nice it would be to have all your servers connected to Ethernet switches with Gigabit Ethernet and all your users using 100BaseT-switched connections. Of course, all your switches would connect with Gigabit links as well. Add xDSL and cable to connect to the Internet and you have more bandwidth than you ever could have imagined just a few years ago. Will it be enough bandwidth a few years from now? Probably not. If you have the bandwidth, users will find a way to use it.
Parkinson’s Law states that data expands to fill the space available for storage, but experience shows that it can be equally applied to bandwidth.
Cisco’s Enterprise model shows a number of different blocks, as defined in Chapter 1, “The Campus Network.” Gigabit Ethernet has value in a number of these different blocks.
The Server Module is a natural choice, because the high demand placed on the network bandwidth by some modern applications would certainly be able to utilize gigabit availability.
The Building Distribution Module carries large amounts of inter-VLAN traffic, and as the 20:80 rule kicks in even more, this additional traffic would benefit from gigabit-speed data transfer.
The Core Module is responsible for connecting all other modules, and it is certain that gigabit throughput would suit the three general principles of core data requirements: speed, speed and more speed!
The Management Module, Building Module, and Edge Distribution Module are less likely at the moment to need gigabit speeds. Most management machines have less data to transfer than applications, most users would be more than satisfied with 100Mbps full-duplex, and the slower WAN speeds at the edge of the network does not need serving by gigabit transfer. Nonetheless, there is rarely such a thing as an average network, and you would be well advised to consider carefully where you might get the best from this exciting technology.
Gigabit Ethernet became an IEEE 802.3 standard in the summer of 1998. The standard was called 802.3z. Gigabit is a combination of Ethernet 802.3 and FiberChannel and uses Ethernet framing the same way 10BaseT and FastEthernet do. This means that not only is it fast, but it can run on the same network as older Ethernet technology, which provides a nice migration plan. The goal of the IEEE 802.3z was to maintain compatibility with the 10Mbps and 100Mbps existing Ethernet network. They needed to provide a seamless operation to forward frames between segments running at different speeds. The committee kept the minimum and maximum frame lengths the same. However, they needed to change the CSMA/CD for half- duplex operation from its 512-bit times to help the distance that Gigabit Ethernet could run.
Will Gigabit ever run to the desktop? Maybe. Probably. People said that FastEthernet would never run to the desktop when it came out, but it’s now common. If Gigabit is run to the desktop, however, it’s hard to imagine what we’ll need to run the backbone with. 10000BaseT to the rescue! Yes, 10 Gigabit Ethernet is out!
In fact, there is now a 10 Gigabit Ethernet Alliance—a group of vendors and other interested parties who together have created the technology behind IEEE 802.3ae, the 10 Gigabit Ethernet standard.
There are some major differences between FastEthernet and Gigabit Ethernet. FastEthernet uses the Media Independent Interface, and Gigabit uses the Gigabit Media Independent Interface (GMII). 10BaseT used the Attachment Unit Interface. A new interface was designed to help FastEthernet scale to 100Mbps, and this interface was redesigned for Gigabit Ethernet. The GMII uses an 8-bit data path instead of the 4-bit path that FastEthernet MII uses. The clocking must operate at 125MHz to achieve the 1Gbps data rate.
Because Ethernet networks are sensitive to the round-trip-delay constraint of CSMA/CD, time slots are extremely important. Remember that in 10BaseT and 100BaseT, the time slots were 512-bit times. However, this is not feasible for Gigabit because the time slot would be only 20 meters in length. To make Gigabit usable on a network, the time slots were extended to 512 bytes (4096-bit times!). However, the operation of full-duplex Ethernet was not changed at all. Table 2.2 compares the new Gigabit Ethernet technologies.
Copper category 5, four-pair wiring, UTP
MMF using 62.5 and 50-micron core, uses a 780-nanometer laser
Single-mode fiber that uses a 9-micron core, 1300-nanometer laser
From 3 kilometers up to 10 kilometers
9-micron single-mode fiber or disposition- shifted fiber
Up to 100 kilometers
If Gigabit Ethernet is used from source to destination, you might consider using Jumbo frames. These are Ethernet frames that are 9000 bytes long. Jumbo frames don’t work well if Gigabit is not used from end to end because fragmentation will take place, causing a small amount of latency. Although Jumbo frames aren’t likely to be used to the desktop, they can speed up the process of data transfer between servers. An e-commerce web server that makes a lot of calls to a database and gets large amounts of data at once would be a good candidate.