3-6 Tutorial on LAN Switching
THOMAS M. HADLEY
All LANs originally consisted of multiple devices communicating with each other on a shared segment. Information appeared at each adapter, which acted only on those frames addressed to it. Any given user had available only part of the 4, 10, or 16M bps that all the other users on the same LAN segment were not using. In the early days, this was no problem, and network response time was instantaneous. Since LANs have clock rates so much higher than wide area networks, some LAN protocols were developed that used far more overhead than necessary. As more people made use of LANs the network became the bottleneck.
The first solution was to look at increasing the LAN speed. This is one of the reasons that Token Ring has long supported either 4 or 16M bps, and why FDDI and 100M-bps Ethernet were developed. While these higher speed solutions improved response time, they require significant LAN infrastructure change new hubs, adapters, and wiring. This led people to look for alternatives.
HOW DO LAN SWITCHES INCREASE LAN SPEED?
Strictly speaking, LAN switches do not increase LAN speed. An easy solution to an overloaded LAN segment is to decrease the number of devices sharing that segment. Once divided, how is the lost connectivity restored? By connecting the segments with bridges (see Exhibit 3-6-1). Most bridges connect only two LAN segments, so as the number of segments increases, this becomes an expensive solution. After years of bridge experience, hardware designers figured out how to better manage the key bridge functions and to build devices that had many ports.
Exhibit 3-6-1. Segmenting the LAN
Since switches are very fast multiport bridges, they could have been called bridges. Bridges, however, had fallen out of favor to routers, so the original developers (or their marketers) decided to call them switches. With a low price per port and easy, drop-in installation, switches were an immediate hit. Today many different LAN switches are offered by each of the companies in the LAN business, as well as by a number of small, startup companies.
LAN SWITCH BASICS
Every frame on a LAN contains a six byte destination address (where it is going) and a six byte source address (where it came from). The destination address can either be for a specific device adapter, or it can be a broadcast (or multicast) address, destined for every device (or some subset) on the LAN. When the frame arrives at the switch port the destination address is looked up in a table to determine which port of the switch the destination device is attached to, and the frame is forwarded to that port and on to the destination (see Exhibit 3-6-2).
Exhibit 3-6-2. Ethernet LAN Switching
Every communication protocol (at least all the common ones) always transmits when it first joins the network, before it expects to receive a frame. Therefore, the switch can receive this transmitted frame, look up the source address in its table, and if it is a new address, add the address and the port it came from to the master table. Thus, when any device attached to any port on the switch sends a frame, the port number of the destination device is already known and in the master table. What happens if a device address is not in the table? Some switches send the frame to every switch port. Others discard any frame whose destination address is unknown.
LAN switches use one of two internal designs. One is to receive incoming frames from each port into a shared memory, and to use a fast processor to look up the destination address in a shared address table, determine which port is associated with that address, and indicate to that output port there is a frame to send and point it to the frames location in shared memory.
The speed of these devices is dependent on the speed of the central processor, and the capacity is determined by the amount of memory for addresses and for frame buffers. Switches using this approach use a fast, industry standard RISC processor. This design is considered to work well in LANs where there is much broadcast or multicast traffic, since the frame memory is shared, and only one copy of the frame exists, independent of the number of ports. NetBIOS and IPX are examples of these communications protocols.
The second is to use a separate processor and memory at each port. In this design the input processor looks up the destination address in its own table to determine the output port, and sends the frame directly to the output port processor. There is a centrally maintained address-port table, which the individual processors refer to for any unknown destination address. Since most communications between LAN user devices are destined for one of a small number of destination devices, the port of the destination address is usually in the input port table.
The speed of switches with this design is determined by the speed and number of individual port processors present, and the capacity is determined by the memory of each processor, the distribution of frames amongst them, and the speed of the internal bus connecting them together. This design uses specialized processors (Application Specific Integrated Circuits or ASICs), optimized for switch use. The size of the LAN for which they work well is determined by the number of addresses present and the size of the port address tables. Because the function is shared by many processors, this design works well in situations where many of the frames are destined for a single end station. TCP/IP, SNA, and DECNet are examples of communication protocols with this characteristic.
Benefit. LAN switches increase the bandwidth available to each LAN user, without requiring any changes to the infrastructure, including visiting or any change in the desktop.
|