- Explain basic switching concepts and the operation of Cisco switches
Switches forward frames based on the Layer 2 Ethernet MAC addresses. These devices receive Ethernet frames transmitted from other devices and dynamically build a MAC address table based on the source MAC address inside those frames. This MAC address table is commonly referred to as a Content Addressable Memory (CAM) table.
These dynamic entries in the CAM table are not permanent, however. After the switch or bridge stops receiving frames from a certain MAC address (this varies, but it’s typically five minutes), the entry is removed from the CAM table to save memory and processor resources. The exceptions to this are static MAC entries that have been manually configured on a portby-port basis for security and control purposes.
When deciding to which port to forward the Ethernet frame, a switch consults this CAM table and forwards the Ethernet frame based on the destination MAC address of the Ethernet header. In instances where the destination MAC address is not in the table, it copies and forwards the frame out every port except the one at which it was received. This action is commonly known as flooding.
It is important to remember that switches build their MAC address tables using the source address in an Ethernet frame header. In addition, they base their forwarding decisions on the destination MAC address in an Ethernet frame header.
Recall that switches segment LANs into collision domains; however, they still are in a single broadcast domain. Switches do not have entries for broadcast addresses (FFFF.FFFF.FFFF) or multicast addresses (0100.5E00.000-0100.5E7F.FFFF) in their CAM tables. As previously mentioned, when a bridge or a switch receives a frame with a destination MAC address not in its table, it floods that frame out every port.
When a switch receives an Ethernet frame with a broadcast, multicast, or unknown unicast (destina tion MAC address of Ethernet frame not in CAM table), it floods that frame out every port except the one at which it was received.
For instance, consider the switched topology example illustrated in Figure 13.1. When Computers A, B, and C and Printer D originally sent an Ethernet frame, the switch recorded the source MAC address of that frame and the associated port in its CAM table. If Computer A sends an Ethernet frame destined for Printer D’s MAC address of 1111.2222.3333, the switch forwards only that frame out to its Fast Ethernet 0/14 interface. If Computer A sends a broadcast with a destination of FFFF.FFFF.FFFF, that entry does not exist in the CAM table, so that frame is flooded out all interfaces except for Fast Ethernet 0/1.
Notice that Computers B and C are plugged into a hub. So what happens when Computer B sends an Ethernet frame to Computer C? The frame hits the Layer 1 hub, which regenerates the signal out all ports except the one it came in on (regardless of the MAC address because it is a Physical layer device). When the frame reaches the switch, the switch realizes that the source and destination MAC addresses reside on the same interface, so it does not send that frame on to any other ports. This process is also commonly known as filtering.
Frame Transmission Methods
Switches are often classified based on the method in which they process and forward frames in and out of their interfaces. This classification differs depending on the device’s processing capabilities and manufacturer. The three transmission methods that a bridge or switch may use are discussed in the following sections.
Properly named, the store-and-forward method of frame transmission involves the switch, which buffers (stores temporarily in a small memory location) the entire Ethernet frame and performs a cyclic redundancy check (CRC) of that frame to make sure it is not a corrupted frame (damaged or abnormally changed in the frame’s transmission). If the frame calculation detects a bad frame, it is dropped at that point. Thus, the frame is forwarded only if the CRC
calculation results in a normal frame.
Because the entire frame is checked, store-and-forward switching is said to be latency (delay) varying. In other words, depending on the payload inside the frame, the switch takes varying processing times to buffer the entire frame and perform the CRC before sending it to its destination. Although this method sounds like a lengthy process, this is the most widely used
method of switching in Cisco Catalyst switches because the hardware processors for the interfaces are so advanced and robust that the switch hardly works up a sweat.
Cut-through transmissions are practically the antithesis of store-and-forward frame transmission. In fact, instead of processing the entire frame, cut-through switching entails the switch buffering just enough information to know where to forward the frame before sending it on to another segment. In other words, it looks only up to the destination MAC address in the Ethernet header and sends it on regardless of whether the frame contains errors.
This hot-potato method of frame transmission was once appealing for devices with low processing power. Because it has to inspect only the beginning of an Ethernet frame header, latency is not a factor with this method. The downside of cut-through switching, however, is that it still passes bad frames on to other segments because it does not perform CRC calculations
of any kind.
In a true Goldilocks fashion, if cut-through is too hot and store-and-forward is too cold, fragment-free may be just right for you. Fragment-free is a hybrid of the two transmission methods because it buffers up to the first 64 bytes of a frame (all collisions occur within the first 64 bytes). This obviously is not as fast as cut-through; nevertheless, it ensures that many of the invalid frames in the LAN are not transmitted on to other segments. Figure 13.2 illustrates how much of an Ethernet frame is buffered and processed with each of the three transmission methods discussed.
Half- and Full-Duplex Connections
Data communication on switch ports can occur in either half- or full-duplex transmissions. Half-duplex connections are unidirectional in that data can be sent in only one direction at a time. This is similar to two-way radios or walkie-talkies, in which only one person can speak at one time. With half-duplex communication in an Ethernet network, CSMA/CD (carrier sense multiple access with collision detection) is enabled, which results in 50 to 60% of the bandwidth on the link being available to be used.
Full duplex, on the other hand, is indicative of two-way communication in which devices can send and receive information at the same time. With these connections, CSMA/CD is automatically disabled, allowing for theoretically 100% of the bandwidth in both directions. In fact, it uses the two wires that typically are used for detecting collisions to simultaneously transmit and receive. Because CSMA/CD is disabled, that means the connection has to be in an environment where collisions cannot occur. In other words, it must be connected to a switch or directly connected with a cross-over cable.
If you are connected to a hub, the connection must be half-duplex with CSMA/CD. When running full duplex, you must be directly connected to a switch.
Because full duplex allows 100% in both directions, it is sometimes advertised at twice the speed. For instance, a 100Mbps interface might be marketed as achieving 200Mbps. Although it is advertised as 200Mbps, in reality, you are receiving 200Mbps of throughput.
You have already seen how switches operate when connected to end-user devices such as PCs, printers, and servers. However, when switches are connected to other switches to form a redundant network, a switching loop can occur. Figure 13.3 illustrates a scenario in which a switching loop can occur.
In this design, redundant links interconnect the switches. Although it is a good idea to have redundancy in the network, the problem arises when a computer sends out a frame with a broadcast, multicast, or unknown unicast destination MAC address. Recall that any of these three transmissions causes a switch to copy and flood that frame out all ports except for the one on which it came in. So if Computer A sends a broadcast, Switch A floods that out to Switches B and D. Again, if this is a broadcast message, Switches B and D flood that frame out to Switch C. Staying true to its design, Switch C floods the frame back to Switches B and D, and so on. Broadcasts continuously circle the switched network until ultimately the amount of broadcast traffic consumes the switched network’s bandwidth and all traffic ceases to flow. This unsettling scenario just described is called a broadcast storm and can be avoided completely by using a Layer 2 protocol sent among switches called the Spanning Tree Protocol.