Network Switching (Part II)

The Pieces of Technology

In 1980, a group of vendors consisting of Digital Equipment Corporation (DEC), Intel, and Xerox created what was known as the DIX standard. Ultimately, after a few modifications, it became the IEEE 802.3 standard. It is the 802.3 standard that most people associate with the term Ethernet.The Ethernet networking technology was invented by Robert M. Metcalfe while he was working at the Xerox Palo Alto Research Center in the early 1970s. It was originally designed to help support research on the “office of the future.” At first, the network’s speed was limited to 3Mbps.Ethernet is a multiaccess, packet−switched system with very democratic principles. The stations themselves provide access to the network, and all devices on an Ethernet LAN can access the LAN at any time. Ethernet signals are transmitted serially, one bit at a time, over a shared channel available to every attached station. To reduce the likelihood of multiple stations transmitting at the same time, Ethernet LANs use a mechanism known as Carrier Sense Multiple Access Collision Detection (CSMA/CD) to listen to the network and see if it is in use. If a station has data to transmit, and the network is not in use, the station sends the data. If two stations transmit at the same time, a collision occurs. The stations are notified of this event, and they instantly reschedule their transmissions using a specially designed back−off algorithm. As part of this algorithm, each station involved chooses a random time interval to schedule the retransmission of the frame. In effect, this process keeps the stations from making transmission attempts at the same time and prevents a collision.After each frame transmission, all stations on the network contend equally for the next frame transmission. This competition allows access to the network channel in a fair manner. It also ensures that no single station can lock out the other stations from accessing the network. Access to the shared channel is determined by theMedia Access Control (MAC) mechanism on each Network Interface Card (NIC) located in each network node. The MAC address uses a physical address which, in terms of the OSI Reference Model, contains the lowest level address. This is the address used by a switch. The router at Layer 3 uses a protocol address, which is referred as a logical address.CSMA/CD is the tool that allows collisions to be detected. Each collision of frames on the network reduces the amount of network bandwidth that can be used to send information across the physical wire. CSMA/CD also forces every device on the network to analyze each individual frame and determine if the device was the intended recipient of the packet. The process of decoding and analyzing each individual packet generates additional CPU usage on each machine, which degrades each machine’s performance.As networks grew in popularity, they also began to grow in size and complexity. For the most part, networks began as small isolated islands of computers. In many of the early environments, the network was installed over a weekend—when you came in on Monday, a fat orange cable was threaded throughout the organization, connecting all the devices. A method of connecting these segments had to be derived. In the next few sections, we will look at a number of approaches by which networks can be connected. We will look at repeaters, hubs, bridges, and routers, and demonstrate the benefits and drawbacks to each approach.

Repeaters

The first LANs were designed using thick coaxial cables, with each station physically tapping into the cable. In order to extend the distance and overcome other limitations on this type of installation, a device known as a repeater is used. Essentially, a repeater consists of a pair of back−to−back transceivers. The transmit wire on one transceiver is hooked to the receive wire on the other, so that bits received by one transceiver are immediately retransmitted by the other.Repeaters work by regenerating the signals from one segment to another, and they allow networks to overcome distance limitations and other factors. Repeaters amplify the signal to further transmit it on the segment because there is a loss in signal energy caused by the length of the cabling. When data travels through the physical cable it loses strength the further it travels. This loss of the signal strength is referred to as attenuation.These devices do not create separate networks; instead, they simply extend an existing one. A standard rule of thumb is that no more than three repeaters may be located between any two stations. This is often referred to as the 5−4−3 rule, which states that no more than 5 segments may be attached by no more than 4 repeaters, with no more than 3 segments populated with workstations. This limitation prevents propagation delay, which is the time it takes for the packet to go from the beginning of the link to the opposite end.As you can imagine, in the early LANs this method resulted in a host of performance and fault−isolation problems. As LANs multiplied, a more structured approach called 10BaseT was introduced. This method consists of attaching all the devices to a hub in the wiring closet. All stations are connected in a point−to−point configuration between the interface and the hub.

Hubs

A hub, also known as a concentrator, is a device containing a grouping of repeaters. Similar to repeaters, hubs are found at the Physical layer of the OSI Model. These devices simply collect and retransmit bits. Hubs are used to connect multiple cable runs in a star−wired network topology into a single network. This design is similar to the spokes of a wheel converging on the center of the wheel.Many benefits derive from this type of setup, such as allowing interdepartmental connections between hubs, extending the maximum distance between any pair of nodes on the network, and improving the ability to isolate problems from the rest of the network.Six types of hubs are found in the network:

• Active hubs—Act as repeaters and eliminate attenuation by amplifying the signals they replicate to all the attached ports.
• Backbone hubs—Collect other hubs into a single collection point. This type of design is also knownas a multitiered design. In a typical setup, servers and other critical devices are on high−speed FastEthernet or Gigabit uplinks. This setup creates a very fast connection to the servers that thelower−speed networks can use to prevent the server or the path to the server from being a bottleneckin the network.
• Intelligent hubs—Contain logic circuits that shut down a port if the traffic indicates that malformedframes are the rule rather than the exception.
• Managed hubs—Have Application layer software installed so that they can be remotely managed.Network management software is very popular in organizations that have staff responsible for anetwork spread over multiple buildings.
• Passive hubs—Aid in producing attenuation. They do not amplify the signals they replicate to all theattached ports. These are the opposite of active hubs.
• Stackable hubs—Have a cable to connect hubs that are in the same location without requiring the datato pass through multiple hubs. This setup is commonly referred to as daisy chaining.
  

In all of these types of hub configurations, one crucial problem exists: All stations share the bandwidth, andthey all remain in the same collision domain. As a result, whenever two or more stations transmit simultaneously on any hub, there is a strong likelihood that a collision will occur. These collisions lead tocongestion during high−traffic loads. As the number of stations increases, each station gets a smaller portionof the LAN bandwidth. Hubs do not provide microsegmentation and leave only one collision domain.

Bridges

A bridge is a relatively simple device consisting of a pair of interfaces with some packet buffering and simplelogic. The bridge receives a packet on one interface, stores it in a buffer, and immediately queues it fortransmission by the other interface. The two cables each experience collisions, but collisions on one cable donot cause collisions on the other. The cables are in separate collision domains.Note Some bridges are capable of connecting dissimilar topologies.The term bridging refers to a technology in which a device known as a bridge connects two or more LANsegments. Bridges are OSI Data Link layer, or Layer 2, devices that were originally designed to connect twonetwork segments. Multiport bridges were introduced later to connect more than two network segments, andthey are still in use in many networks today. These devices analyze the frames as they come in and makeforwarding decisions based on information in the frames themselves.To do its job effectively, a bridge provides three separate functions:

• Filtering the frames that the bridge receives to determine if the frame should be forwarded
  
• Forwarding the frames that need to be forwarded to the proper interface
  
• Eliminating attenuation by amplifying received data signals

Bridges learn the location of the network stations without any intervention from a network administrator or any manual configuration of the bridge software. This process is commonly referred to as self−learning. When a bridge is turned on and begins to operate, it examines the MAC addresses located in the headers of frames passed through the network. As the traffic passes through the bridge, the bridge builds a table of known source addresses, assuming the port from which the bridge received the frame is the port to which the device is a sending device is attached.In this table, an entry exists that contains the MAC address of each node along with the bridge interface and port on which it resides. If the bridge knows that the destination is on the same segment as the source, it drops the packet because there is no need to transmit it. If the bridge knows that the destination is on another segment, it transmits the packet on that segment or port to that segment only. If the bridge does not know the destination segment, the bridge transmits a copy of the frame to all the interface ports in the source segment using a technique known as flooding. For each packet an interface receives, the bridge stores in its table the following information:

• The frame’s source address
  
• The interface the frame arrived on
  
• The time at which the switch port received the source address and entered it into the switching table
  

Note Bridges and switches are logically equivalent.

There are four kinds of bridges:

• Transparent bridge—Primarily used in Ethernet environments. They are called transparent bridgesbecause their presence and operation are transparent to network hosts. Transparent bridges learn andforward packets in the manner described earlier.
  
• Source−route bridge—Primarily used in Token Ring environments. They are called source−routebridges because they assume that the complete source−to−destination route is placed in frames sentby the source.
  
• Translational bridge—Translators between different media types, such as Token Ring and Ethernet.
  
• Source−route transparent bridge—A combination of transparent bridging and source−route bridgingthat enables communication in mixed Ethernet and Token Ring environments.

Broadcasts are the biggest problem with bridges. Some bridges help reduce network traffic by filtering packets and allowing them to be forwarded only if needed. Bridges also forward broadcasts to devices on all segments of the network. As networks grow, so does broadcast traffic. Instead of frames being broadcast through a limited number of devices, bridges often allow hundreds of devices on multiple segments to broadcast data to all the devices. As a result, all devices on all segments of the network are now processing data intended for one device. Excessive broadcasts reduce the amount of bandwidth available to end users.This situation causes bandwidth problems called network broadcast storms. Broadcast storms occur when broadcasts throughout the LAN use up all available bandwidth, thus grinding the network to a halt. Network performance is most often affected by three types of broadcast traffic: inquiries about the availability of a device, advertisements for a component’s status on the network, and inquiries from one device trying to locate another device. The following are the typical types of network broadcasts:

• Address Resolution Protocol (ARP)
  
• Internetwork Packet Exchange (IPX)
  
• Get Nearest Server (GNS) requests
  
• IPX Service Advertising Protocol (SAP)
  
• Multicast traffic broadcasts
  
• NetBIOS name requests
  

These broadcasts are built into the network protocols and are essential to the operation of the network devices using these protocols.Due to the overhead involved in forwarding packets, bridges also introduce a delay in forwarding traffic. This delay is known as latency. Latency delay is measured from the moment a packet enters the input port on the switch until the time the bridge forwards the packet out the exit port. Bridges can introduce 20 to 30 percent loss of throughput for some applications. Latency is a big problem with some timing−dependent technologies, such as mainframe connectivity, video, or voice.High levels of latency can result in loss of connections and noticeable video and voice degradation. The inherent problems of bridging over multiple segments including those of different LAN types with Layer 2 devices became a problem to network administrators. To overcome these issues, a device called a router, operating at OSI Layer 3, was introduced.

Routers

Routers are devices that operate at Layer 3 of the OSI Model. Routers can be used to connect more than one Ethernet segment with or without bridging. Routers perform the same basic functions as bridges and also forward information and filter broadcasts between multiple segments. Figure 1.2 shows routers segmenting multiple network segments. Using an OSI network Layer 3 solution, routers logically segment traffic intosubnets.

Figure 1.2: Routers connecting multiple segments.

Routers were originally introduced to connect dissimilar network media types as well as to provide a means to route traffic, filter broadcasts across multiple segments, and improve overall performance. This approach eliminated broadcasts over multiple segments by filtering broadcasts. However, routers became a bottleneck in some networks and also resulted in a loss of throughput for some types of traffic.When you are connecting large networks, or when you are connecting networks to a WAN, routers are very important. Routers will perform media conversion, adjusting the data link protocol as necessary. With a router, as well as with some bridges, you can connect an Ethernet network and a Token Ring network. Routers do have some disadvantages. The cost of routers is very high, so they are an expensive way to segment networks. If protocol routing is necessary, you must pay this cost. Routers are also difficult to configure and maintain, meaning that you will have a difficult time keeping the network up and running.Knowledgeable workers who understand routing can be expensive.Routers are also somewhat limited in their performance, especially in the areas of latency and forwarding rates. Routers add about 40 percent additional latency from the time packets arrive at the router to the time they exit the router. Higher latency is primarily due to the fact that routing requires more packet assembly and disassembly. These disadvantages force network administrators to look elsewhere when designing many large network installations.

Switches

A new option had to be developed to overcome the problems associated with bridges and routers. These new devices were called switches. The term switching was originally applied to packet−switch technologies, such as Link Access Procedure, Balanced (LAPB); Frame Relay; Switched Multimegabit Data Service (SMDS); and X.25. Today, switching is more commonly associated with LAN switching and refers to a technology that is similar to a bridge in many ways.Switches allow fast data transfers without introducing the latency typically associated with bridging. They create a one−to−one dedicated network segment for each device on the network and interconnect these segments by using an extremely fast, high−capacity infrastructure that provides optimal transport of data on a LAN; this structure is commonly referred to as a backplane. This setup reduces competition for bandwidth on the network, allows maximum utilization of the network, and increases flexibility for network designers and implementers.Ethernet switches provide a number of enhancements over shared networks. Among the most important is microsegmentation, which is the ability to divide networks into smaller and faster segments that can operate at the maximum possible speed of the wire (also known as wire−speed).To improving network performance, switches must address three issues:

• They must stop unneeded traffic from crossing network segments.
• They must allow multiple communication paths between segments.
• They cannot introduce performance degradation.

Routers are also used to improve performance. Routers are typically attached to switches to connect multiple LAN segments. A switch forwards the traffic to the port on the switch to which the destination device is connected, which in turn reduces the traffic to the other devices on the network. Information from the sending device is routed directly to the receiving device. No device other than the router, switch, and end nodes sees or processes the information.The network now becomes less saturated, more secure, and more efficient at processing information, and precious processor time is freed on the local devices. Routers today are typically placed at the edge of the network and are used to connect WANs, filter traffic, and provide security. See Figure 1.3.

Figure 1.3: Routers and switches

Like bridges, switches perform at OSI Layer 2 by examining the packets and building a forwarding table based on what they hear. Switches differ from bridges by helping to meet the following needs for network designers and administrators:

• Provide deterministic paths
• Relieve network bottlenecks
• Provide deterministic failover for redundancy
• Allow scalable network growth
• Provide fast convergence
• Act as a means to centralize applications and servers
• Have the capacity to reduce latency