Switch Hardware Types

It's not much of a stretch to argue that switches are a network administrator's dream come true, at least most of the time. But when you decide to incorporate switches, there is something else to consider: Not all switches use the same technology. The importance of this distinction depends on which of these functions of a switch is the most important to you:

• Increasing the bandwidth for computers attached to a switch.

• Decreasing the possibility that frame errors will be propagated end-to-end in a network link.

Many architectures are used for switching, as described in the following sections, and because of that, many approaches have been and are being tried. Some involve software that makes decisions much like a router and sends frames on their way. Others are hardware-based and can perform much better because no single component, such as a CPU, can be bogged down when too much traffic passes through the switch. Two basic modes of operation can be used by a switch when it forwards a packet out of a selected port: cut-through mode and store-and-forward mode.

Cut-Through Switches

A cut-through switch begins transmitting the incoming frame on the outgoing port after it receives the header information, or about 20 or 30 bytes. All the switch needs to determine on which port to output the frame is the destination address (hardware address), which is determined by the MAC address found in the frame header. The switch continues to receive information and transmit it until the frame has been "switched" from one port to another. The advantage to this mode of operation is speed. As long as nothing else goes wrong, the packet continues on to its destination at a fast pace with little time involved in the switch. The switch is said to be switching at wire speed. That is, the delay introduced by the switching function is so insignificant that to the end workstations, the full bandwidth is available for use.

This method has several disadvantages, however. The switch begins to send the packet out before it knows whether the frame is damaged in any way. If the frame has corrupted data, the switch won't be able to detect it unless it first receives the entire frame and then computes the CRC (cyclic redundancy check) value stored in the frame check sequence field. If a frame is badly malformed, as when an NIC sends out a frame that is too long, a cut-through switch might think it is a broadcast packet and send it out of all ports, causing unnecessary traffic congestion.

Store-and-Forward Switches

In the store-and-forward switch, the switch buffers the frame in its own memory before beginning to send it out of the appropriate port. This technique boasts two main advantages:

• The switch can connect two different topologies, such as 10Mbps and 100Mbps networks, without having to worry about the different speeds.

• The switch can operate like a bridge and check the integrity of the frame, allowing it to discard damaged frames and not propagate them onto other network segments. This means that a malformed frame received from a local port can be discarded immediately, instead of being sent through the entire switched network until the end-node discovers that an error has occurred.

Although the store-and-forward technology increases the latency factor, this delay usually is not a big concern when you consider the increased throughput you can achieve with a switch.

Layer 3 Switches

Just as switches are on an evolutionary upgrade path from hubs and bridges, an enhanced breed of networking device is becoming increasingly popular in large networks. Layer 3 of the OSI model is the Network layer, on which higher-level protocol addresses are introduced into the network. Generally, switches are deployed in a LAN, whereas routers, which use layer 3 addresses (such as an IP address), are used to connect LANs that are separated by some distance, such as in a campus LAN, or to connect WANs. The main difference here is that the switch must examine only a small amount of the frame header to determine the hardware address of a frame and then send the frame out of the correct port. Routers, however, need to dig further into the packet to find the higher-level protocol address, such as an IP address. Routers also must modify the frame header, substituting the router's MAC address as the source address of the frame, examining and modifying the TTL field in the packet and performing checksum calculations to ensure the integrity of the packet. Because of the extra processing involved, routers generally operate at a lower speed than do switches.

Standard routers operating at slower speeds than switches tend to become bottlenecks in a network. To solve this problem, layer 3 switching devices usually take a different approach to the functions a router performs. Routers are like computers (indeed, sometimes a computer with multiple network adapters is used for routing in a small network), and a processor must examine each packet and perform all the functions just mentioned. Layer 3 switches usually implement these functions in application-specific integrated circuits (ASICs). By implementing these functions in hardware, some layer 3 switches can operate at just about wire speed, which ordinary routers cannot do.

Some layer 3 switches use proprietary technologies, because standards are not complete for this type of device at this time. Whatever method they use, the idea is to identify streams of traffic that are all traveling to the same destination, and output them on the appropriate port as fast as possible.

Most products that advertise themselves as layer 3 switches also function as routers. Layer 3 switching is employed for traffic streams that are easily identifiable. For small traffic loads, the device operates much like a router. In the next few years, you can expect to see layer 3 switching come down in price, making it feasible in smaller networks. For now, however, the cost might not justify the increase in speed you will achieve. For example, if a router is a bottleneck in your network that sits between client computers and servers, consider moving servers closer to the clients so that the network traffic flow doesn't have to pass through the router.

A true Layer 3 switch should support most of the following features:

• Support for TCP/IP as well as other protocols such as SNA, XNS, AppleTalk and IPX; this is important if you use other network protocols

• Multicast control for broadcasting streaming video and audio

• SNMP support for network and switch management

• IEEE 802.1D spanning tree protocol support

• IEEE 802.1Q VLAN support

• Port trunking to provide automatic swichover to parallel backbone connections

• IEEE 802.3x full-duplex flow control support

• Fault tolerance features such as hot-swapping, multiple fans and power supplies, multiple CPUs

Some Layer 3 switches also support switching Layer 4 and higher layers. This provides for better quality-of-service support and traffic control.

Another interesting development in routing technologies, called Multi-Label Protocol Switching, is discussed in Chapter 33, "Routing Protocols." This method of wire-speed switching, generally found in high-end Internet core routers, is defined by RFC documents, which are either proposed standards or informational documents. Here are some of them:

• RFC 3034, "Use of Label Switching on Frame Relay Networks Specification"

• RFC 3270, "Multi-Protocol Label Switching (MPLS) Support of Differentiated Services"

• RFC 3468, "The Multi-Protocol Label Switching (MPLS) Working Group Decision on MPLS Signaling Protocols"

• RFC 3471, "Generalized Multi-Protocol Label Switching (GMPLS) Signaling Functional Description"

Putting a Switch in Your Home Office

Switches, similar to hubs, come in all sizes and shapes. As stated at the beginning of this chapter, the switch has replaced the hub for all practical purposes. There is no longer a major cost difference between switches and hubs. In fact, it can be rather difficult to find new hubs. On a SOHO network, a wired or wireless router with an integrated switch is an ideal solution. It can be expanded to handle more client devices by connecting an external switch or WAP.

Installing a switch of this sort requires very little effort. You basically plug the network cables from your computers into the ports on the back of the switch and then power up the switch. If you expect your network to grow during the next year or two, you should know that most switches have an "uplink port" also. If this is the case, the documentation for your switch will point out which port is used for this function. The uplink port is used to attach your switch to another switch should your network grow and you need additional ports to connect the new computers. If your switch doesn't have an uplink port, you can use a cross-over cable to connect two standard switch ports to achieve the same result. A cross-over cable basically just swaps the transmit and receive wires so that the ports can communicate. Additionally, some uplink ports can be converted to a regular port so that you can attach a computer instead. There is usually a button or switch that can perform this function. Check the documentation!

Stackable and Chassis Switches

For larger networks, you'll find that switches come in stackable and chassis models. Stackable switches have an interconnect port you can use to link them together, so you can add capacity as your network grows. Chassis switches fit a lot of switching capacity into a very small space, providing a large number of ports. Chassis switches can be placed into computer racks and take up much less room than other types of switches. The term "blade" has come into vogue recently to describe servers, switches, and other devices that can be located in a densely populated computer rack. These kinds of switches also provide other functions, such as better management capabilities, support for the Simple Network Management Protocol (SNMP) and Remote Monitoring (RMON), and the capability to create virtual LANs, which is the subject of the next chapter.