Maximizing the Benefits of Switching
As devices are added to LANs to accommodate more users, and more bandwidth is required by more networked software applications, maintaining an acceptable level of network performance becomes an increasing challenge. There are a number of ways to enhance switched Ethernet LANs to meet the demands of users for performance and availability.
Microsegmentation
Microsegmentation eliminates the possibility of collisions on the network segment, providing a number of benefits in increasing network performance. Figure 2-21 shows how microsegmentation is accomplished using a switch.
Figure 2-21 Microsegmentation
Implementing LAN switching provides microsegmentation. Each device on a network segment is connected directly to a switch port and does not have to compete with any other device on the segment for bandwidth. This important function eliminates collisions and increases the effective data rate through full-duplex operation, resulting in a significant increase in available bandwidth.
Example: Getting a Dedicated On-Ramp
Data transmission can be compared to a freeway, with data frames traveling over the freeway like automobiles. Just as automobiles use on-ramps to access the freeway, devices join the network when they want to transmit data. As more and more cars travel on the freeway, however, the on-ramps can become congested, allowing access to only a few cars, and there can even be collisions. If each car had its own on-ramp, however, all the cars would have equal access to the freeway, and there would be no delays or collisions. The microsegmentation that LAN switches provide gives each network device its own “onramp” so that the device does not have to compete with other devices to use the network “freeway.”
Duplex Communication
Full-duplex communication increases effective bandwidth by allowing both ends of the connection to transmit simultaneously. However, this method of optimizing network performance requires microsegmentation before full-duplex communication can occur.
Half-duplex transmission mode implements Ethernet carrier sense multiple access collision detect (CSMA/CD). The traditional shared LAN operates in half-duplex mode, like with hubs, and is susceptible to transmission collisions across the wire.
Full-duplex Ethernet significantly improves network performance without the expense of installing new media. Full-duplex transmission between stations is achieved by using pointto-point Ethernet, Fast Ethernet, and Gigabit Ethernet connections. This arrangement is collision-free. Frames sent by the two connected end nodes cannot collide because the end nodes use two separate circuits in the unshielded twisted-pair (UTP) cable. Each fullduplex connection uses only one port.
Full-duplex port connections are point-to-point links between switches or end nodes, but not between shared hubs. Nodes that are directly attached to a dedicated switch port with network interface cards (NIC) that support full-duplex should be connected to switch ports that are configured to operate in full-duplex mode. Most Ethernet, Fast Ethernet, and Gigabit Ethernet NICs sold today offer full-duplex capability. In full-duplex mode, the collision detect circuit is disabled.
Nodes that are attached to hubs that share their connection to a switch port must operate in half-duplex mode because the end stations must be able to detect collisions.
Figure 2-22 shows how full-duplex can be implemented for bidirectional communication on a switch where it connects to a host but this feature cannot be configured for connectivity for a hub.
Figure 2-22 Full- and Half-Duplex Connections
Standard shared Ethernet configuration efficiency is typically rated at 50 to 60 percent of the 10-Mbps bandwidth. Full-duplex Ethernet offers 100 percent efficiency in both directions (10-Mbps transmit and 10-Mbps receive).
Full-Duplex Communication
Because each device on a microsegmented switched LAN is connected directly to a port on a switch, the switch port and that device have a point-to-point connection. In networks with hubs instead of switches, devices can communicate in only one direction at a time because they must compete for the network bandwidth. This type of communication is referred to as half-duplex communication, because it allows data to be either sent or received at one time, but not both. Microsegmented switch ports, however, can provide the devices connected to them with full-duplex-mode communication, allowing the devices to both send and receive data simultaneously. This ability effectively doubles the amount of bandwidth between the devices.
Example: Data Conversations
If you use a voice communication device such as a walkie-talkie, you will be communicating in half-duplex mode. You can talk, but then you must stop talking to hear what the person on the other end of the line is saying. With a telephone, however, you can communicate with someone in full-duplex mode; each person can both talk and hear what the other person says simultaneously.
Duplex Interface Configuration
Example 2-11 shows how to configure the speed and duplex on a 2960 series switch.
Example 2-11 Configuring Duplex
SwitchX(config)# interface fa0/1
SwitchX(config-if)# duplex {auto | full | half}
SwitchX(config-if)# speed {1 0 | 1 00 | 1 000 | auto}
Use the duplex interface configuration command to specify the duplex mode of operation for switch ports.
The duplex parameters on the Cisco Catalyst 2960 series are as follows:
- auto sets auto-negotiation of duplex mode.
- full sets full-duplex mode.
- half sets half-duplex mode.
For Fast Ethernet and 10/100/1000 ports, the default is auto. For 100BASE-FX ports, the default is full. The 10/100/1000 ports operate in either half-duplex or full-duplex mode when they are set to 10 or 100 Mbps, but when set to 1000 Mbps, they operate only in fullduplex mode.
100BASE-FX ports operate only at 100 Mbps in full-duplex mode.
NOTE To determine the default duplex mode settings for the Gigabit Interface Converter (GBIC) module ports, refer to the documentation that came with your GBIC module.
Example: Showing Duplex Options
Verify the duplex settings by using the show interfaces command, as shown in Example 2-12, on the Catalyst 2960 series. The show interfaces privileged EXEC command displays statistics and status for all or specified interfaces.
Example 2-12 Showing Duplex on an Interface
SwitchX# show interfaces fastethernet0/2 FastEthernet0/2 is up, line protocol is up (connected) Hardware is Fast Ethernet, address is 0008.a445.9b42 (bia 0008.a445.9b42) MTU 1500 bytes, BW 10000 Kbit, DLY 1000 usec, reliability 255/255, txload 1/255, rxload 1/255 Encapsulation ARPA, loopback not set Keepalive set (10 sec) Half-duplex, 10Mb/s input flow-control is unsupported output flow-control is unsupported ARP type: ARPA, ARP Timeout 04:00:00 Last input 00:00:57, output 00:00:01, output hang never Last clearing of “show interface” counters never Input queue: 0/75/0/0 (size/max/drops/flushes); Total output drops: 0 Queueing strategy: fifo Output queue: 0/40 (size/max) 5 minute input rate 0 bits/sec, 0 packets/sec 5 minute output rate 0 bits/sec, 0 packets/sec 323479 packets input, 44931071 bytes, 0 no buffer Received 98960 broadcasts (0 multicast) 1 runts, 0 giants, 0 throttles 1 input errors, 0 CRC, 0 frame, 0 overrun, 0 ignored 0 watchdog, 36374 multicast, 0 pause input 0 input packets with dribble condition detected 1284934 packets output, 103121707 bytes, 0 underruns 0 output errors, 2 collisions, 6 interface resets 0 babbles, 0 late collision, 29 deferred 0 lost carrier, 0 no carrier, 0 PAUSE output 0 output buffer failures, 0 output buffers swapped out
Auto-negotiation can at times produce unpredictable results. Auto-negotiation can happen when an attached device, which does not support auto-negotiation, is operating in fullduplex. By default, the Catalyst switch sets the corresponding switch port to half-duplex
mode. This configuration, half-duplex on one end and full-duplex on the other, causes late collision errors at the half-duplex end. To avoid this situation, manually set the duplex parameters of the switch to match the attached device.
If the switch port is in full-duplex mode and the attached device is in half-duplex mode, check for frame check sequence (FCS) errors on the switch full-duplex port. You can use the show interfaces command to check for FCS late collision errors.
Need for Different Media Rates in an Enterprise Network
Large networks include large numbers of end systems, servers, and network devices, and each can require different speeds to be interconnected. This topic describes the reasons for different speed requirements in an enterprise network.
There are a number of higher-speed Ethernet protocols (such as Fast Ethernet and Gigabit Ethernet) that can provide the speed that is required to ensure the performance that is vital to large networks. However, the cost of implementing high-speed connections in all parts of an enterprise network would be very high, and high-speed connections would not be consistently used by all users and devices. Using a hierarchy of Ethernet connectivity, therefore, is usually the most efficient way to supply speed where it will be most effective.
In a typical connectivity hierarchy, the end-user devices are usually referred to as the “access-level” systems, because they are the primary point at which the network is accessed to transmit data. End-user systems are aggregated at the server or workgroup “distribution” level, and if necessary, end-user systems will use the backbone, or “core”: level, to reach another distribution device. Higher connectivity speed is usually reserved for those devices that transmit large quantities of data from multiple users, notably at the distribution and core levels. This three-tier hierarchy is shown in Figure 2-23.
Physical Redundancy in an Ethernet LAN
When multiple switches are implemented on the same network and when there are multiple redundant physical connections between the switches, there is a potential for intentional or unintentional physical loops. When loops occur, broadcast storms can be created, propagating frames throughout the network in an endless loop.
Figure 2-23 Three-Tier Hierarchy of Connectivity
Adding switches to LANs can add the benefit of redundancy, that is, connecting two switches to the same network segments to ensure continual operations in case there are problems with one of the segments. Redundancy can ensure the availability of the network at all times. However, when switches are used for redundancy in a network, there is the potential problem of loops. When a host on one network segment transmits data to a host on another network segment, and the two are connected by two or more switches, each switch receives the data frames, looks up the location of the receiving device, and forwards the frame. Because each switch forwarded the frame, there is a duplication of each frame.
This process results in a loop, and the frame circulates between the two paths without being removed from the network. The MAC tables might also be updated with incorrect MAC address port mapping information, resulting in inaccurate forwarding. In addition to basic connectivity problems, the proliferation of broadcast messages in networks with loops represents a serious network problem. Because of how switches operate, any multicast, broadcast, or unknown traffic will be flooded out to all ports except the incoming port. The resulting effect is a “broadcast storm” of traffic being looped endlessly through the network, almost instantly consuming the available bandwidth.
Example: Loops in a Switched Network
This looping problem is demonstrated in Figure 2-24.
Figure 2-24 Switching Loops in a Network
Suppose that a host named London sends a frame to a host named Rome. London resides on network segment A, and Rome resides on network segment B. Redundant connections between switches and hosts are provided to ensure continual operations in the case of a segment failure. For the example shown in Figure 2-24, it is assumed that none of the
switches have learned host B’s address.
Switch 1 receives the frame destined for host B and floods it out to switches 2 and 3. Both switch 2 and switch 3 receive the frame from London (through switch 1) and correctly learn that London is on segments 1 and 2, respectively. Each switch forwards the frame to switch 4.
Switch 4 receives two copies of the frame from London, one copy through switch 2 and one copy through switch 3. Assume that the frame from switch 2 arrives first. Switch 4 learns that London resides on segment 3. Because switch 4 does not know Rome’s MAC address, it forwards the frame from switch 2 to Rome and switch 3. When the frame from switch 3 arrives at switch 4, switch 4 updates its table to indicate that London resides on segment 4. It then forwards the frame to Rome and switch 2.
Switches 2 and 3 now change their internal tables to indicate that London is on segments 3 and 4, respectively. If the initial frame from London were a broadcast frame, both switches would forward the frames endlessly, using all available network bandwidth and blocking the transmission of other packets on both segments. This is called a broadcast storm. Loop Resolution with Spanning Tree Protocol (STP)
The solution to loops is STP, which manages the physical paths to given network segments. STP provides physical path redundancy, while preventing the undesirable effects of active loops in the network. Spanning Tree Protocol is on by default in Catalyst switches. Figure 2-25 shows how STP prevents loops by blocking on a redundant path link.
Figure 2-25 STP Prevents Switching Loops
STP behaves as follows:
- STP forces certain ports into a standby state so that they do not listen to, forward, or flood data frames. The overall effect is that even when multiple physical paths exist for redundancy, there is only one active path to each network segment at any given time.
- If there is a problem with connectivity to any of the segments within the network, STP will reestablish connectivity by automatically activating a previously inactive path, if one exists.NOTE Spanning Tree Protocol is covered in further detail in Interconnecting Cisco Networking Devices Part 2 (ICND2).