While the performance of routers has improved significantly in the past few years, any device at Layer 3 must perform additional processing on each packet in order to function. Therefore, the downside of routers is usually their latency and packets-per-second (PPS) performance. Newer routing technologies use network data-flow-based switching and other techniques to route only the first packets and then switch the remainder of the flow.

Bridges and Switches

Switches build upon the same technology as bridges, but during their evolution switches have added features to their offerings. In addition, switches frequently operate at “wire speed,” i.e., any amount of data entering the port will be processed and forwarded without the need to discard the frame. This is a substantial improvement from the first generation of bridges, in which a burst of frames could quickly saturate the buffers.

One of the keys to obtaining performance from a switch is the proper design of the network. Resources, or those devices that service many users, should be provided with the fastest ports available on the switch. Stated another way, it would be poor design to put a file server on a 10MB interface servicing 100MB workstations. The greatest bandwidth should always be allocated to servers and trunk links.

Technically, switches are defined within Layer 2 of the OSI model, and Cisco continues to use this definition. However, as noted in the previous section, modern switches are greatly expanding upon the definition of their original role. For the purposes of this discussion, switches forward frames based only on the MAC layer address.

Switches are also responsible for maintaining VLAN information and may isolate ports based on the end-station MAC address, its Layer 3 address (although forwarding decisions are still based at Layer 2), or the physical port itself.

Most switches operate in one of two forwarding modes. Cut-through switches forward frames as soon as the destination address is seen. No CRC (cyclical redundancy check) is performed, and latency is consistent regardless of frame size. This configuration can permit the forwarding of corrupted frames. The second forwarding mode is called store-and-forward. The entire frame is read into memory, and the CRC is performed before the switch forwards the frame. This prevents corrupted frames from being forwarded, but latency is variable and greater than with cut-through switching.

Although switches are defined in the main text, designers should consider the “real-world” state of the technology. Layer 3 switching routers are capable of handling basic LAN-based Layer 3 functions, including routing and media conversion. Newer switching products are adding Layers 4 and 5 to their forwarding and processing lookups. This high-speed LAN-optimized routing technology is particularly important when considering load-balancing and queuing, because additional information regarding the packet flow can greatly increase the efficiency of the overall network capacity.

Summary of Routing and Switching

This overview of the LAN technologies provides the designer additional information about routing and switching technologies. This information is crucial to understanding the methods for designing scalable networks. Designers should consider the differences in broadcast and collision control and should also take note of loop prevention.

Hubs and repeaters Hubs and repeaters work at Layer 1 of the OSI reference model. No filtering or blocking occurs, and they are used to extend cable length.

Bridges and switches Bridges and switches limit the collision domain but not the broadcast domain. Bridges and switches control loops with the Spanning-Tree Protocol (STP). Switches are considered high-speed, multi-port transparent bridges, with advanced features. These advanced features include broadcast suppression and VLAN trunking. Bridges and switches both operate at Layer 2 (the MAC layer). Switches also incorporate bandwidth flexibility—for example, a LAN using a hub shares all bandwidth among the stations. Thus, 10 stations must contend for a single 10Mbps network. Installation of a switch immediately provides each station with a dedicated 10Mbps, or a total theoretical bandwidth of 100Mbps. The limitation moves to the switch’s backplane and buffers. In the same context, a shared FDDI ring operating at 100Mbps can be replaced with an ATM switch operating at OC-3 speeds (155Mbps). Each port has a dedicated link. Many designers divide shared media by the number of devices—thus, 10 stations on an FDDI ring will each receive 10Mbps. This is a simplified method for estimating performance increases.

Routers Routers operate at Layer 3, limiting the collision and broadcast domains. Loops are handled within the routing protocol, using mechanisms such as split-horizon and time-to-live counters. Routers require logical addressing.

Nodes

Network design can be a precise exercise in which the designer knows exactly how much data will be sent across the network and when these transmissions will occur. Unfortunately, such accuracy would be short-lived and extremely time-consuming to obtain. General guidelines are actually just a means of simplifying the technical process while maintaining sufficient accuracy.

A number of factors combine to determine the number of nodes per network. For example, 10-Base-2 will support only 30 nodes according to the specification, but most installations surpass this threshold. Ignoring this limitation, most network designers today are concerned with Ethernets, broadcasts, and cable distances.