 | My Personal Bookstore & Wishlist |
|
Delivering Data Through Internetworks
Figure 56 shows an internetwork consisting of several networks. The way data are delivered through internetworks involves several topics: • Methods for carrying multiple data streams on common media. • Methods for switching data through paths on the network. • Methods for determining the path to be used. LAN’s generally operate in baseband mode, which means that a given cable is carrying a single data signal at any one time. The various devices on the LAN must take turns using the medium. This generally is a workable approach for LAN’s, because LAN media offer high performance at low cost. Long-distance data communication media are expensive to install and maintain, and it would be inefficient if each media path could support only a single data stream. WAN’s, therefore, tend to use broadband media, which can support two or more data streams. Increasingly, as LAN’s are expected to carry more and different kinds of data, broadband media are being considered for LAN as well. To enable many data streams to share a high-bandwidth medium, a technique called multiplexing is employed.
Figure 57 illustrates one method of time-division multiplexing of digital signals. In figure 57, the signals-carrying capacity of the medium is divided into time slots, with a time slot assigned to each signal, a technique called Time-Division Multiplexing (TMD). Because the sending and receiving devices are synchronised to recognise the same time slots, the receiver can identify each data stream and re-create the original signals. The sending device, which places data into the time slots, is called a multiplexer or mux. The receiving device is called a demultiplexer or demux. TMD can be inefficient. If a data stream falls silent, its time slots are not used and the media bandwidth is under-utilised.
Figure 58 depict a more advanced technique, statistical time-division multiplexing. In figure 58, time slots are still used, but some data streams are allocated more time slots that others. An idle channel, D, is allocated no time slots at all. A device that performs statistical TMD often is called a stat-MUX. On an internetwork, data units must be switched through the various intermediate devices until they are delivered to their destination. Two contrasting methods of switching data are commonly used: Circuit switching and packet switching. Both are used in some form by protocols in common use.
Figure 59 illustrates circuit switching. When two devices negotiate the start of a dialogue, they establish a path, called a circuit, through the network, along with a dedicated bandwidth through the circuit. After establishing the circuit, all data for the dialogue flow through that circuit. The chief disadvantage of circuit switching is that when communication takes place at less than the assigned circuit capacity, bandwidth is wasted. Also, communicating devices can’t take advantage of other, less busy paths through the network unless the circuit is reconfigured. Circuit switching does not necessarily mean that a continuous, physical pathway exists for the sole use of the circuit. The message stream may be multiplexed with other message streams in a broadband circuit. In fact, sharing of media is the more likely case with modern telecommunications. The appearance to the end devices, however, is that the network has configured a circuit dedicated to their use. End devices benefit greatly from circuit switching. Since the path is pre-established, data travel through the network with little processing in transit. And, because multipart messages travel sequentially through the same path, message segments arrive in an order and little effort is required to reconstruct the original message.
Figure 60 illustrates packet switching. Packet switching takes a different and generally more efficient approach to switching data through networks. Messages are broken into sections called packets, which are routed individually through the network. At the receiving device, the packets are reassembled to construct the complete message. Messages are divided into packets to ensure that large messages do not monopolise the network. Packets from several messages can be multiplexed through the same communication channel. Thus, packet switching enables devices to share the total network bandwidth efficiently.
• Datagram services treat each packet as an independent message. The packets, also called datagrams, are routed through the network using the most efficient route currently available, enabling the switches to bypass busy segments and use under-utilised segments. Datagrams frequently are employed on LAN’s and network layer protocols are responsible for routing the datagrams to the appropriate destination. Datagram service is called unreliable, not because it is inherently flawed but because it does not guarantee delivery of data. Recovery of errors is left to upper-layer protocols. Also, if several messages are required to construct a complete message, upper-layer protocols are responsible for reassembling the datagrams in order. Protocols that provide datagram service are called connectionless protocols. • Virtual circuits establish a formal connection between two devices, giving the appearance of a dedicated circuit between the devices. When the connection is established, issues such as messages size, buffer capacities, and network paths are considered and mutually agreeable communication parameters are selected. A virtual circuit defines a connection, a communication path through the network, and remains in effect as the devices remain in communication. This path functions as a logical connection between the devices. When communication is over, a formal procedure releases the virtual circuit. Because virtual circuit service guarantees delivery of data, it provides reliable delivery service. Upper-layer protocols need not be concerned with error detection and recovery. Protocols associated with virtual circuits are called connection-oriented. Bridges, Routers, and Switches:
• The physical address of the destination device, found at the data link layer. Devices that forward messages based on physical addresses generally are called bridges. • The address of the destination network, found at the network layer. Devices that use network addresses to forward messages usually are called routers, although the original name, still commonly used in the TCP/IP world, is gateway. • The circuit that has been established for a particular connection. Devices that route messages based on assigned circuits are called switches.
Figure 61 illustrates the protocol stack model for bridging in terms of the OSI Reference Model. Bridges build and maintain a database that lists known addresses of devices and how to reach those devices. When it receives a frame, the switch consults its database to determine which of its connections should be used to forward the frame. A bridge must implement both the physical and data link layers of the protocol stack. Bridges are fairly simple devices. The receive frames from on connection and forward them to another connection known to be en route to the destination. When more than one route is possible, bridges ordinarily can’t determine which route is most efficient. In fact, when multiple routes are available, bridging can result in frames simply travelling in circles. Having multiple paths available on the network is desirable, however, so that a failure of one path does not stop the network. With Ethernet, a technique called the spanning-tree algorithm enables bridged networks to contain redundant paths. Token Ring uses a different approach to bridging. When a device needs to send to another device, it goes through a discovery process to determine a route to the destination. The routing information is stored in each frame transmitted and is used by bridges to forward the frames to the appropriate networks. Although this actually is a data link layer function, the technique Token Ring uses is called source routing. The bridge must implement two protocol stacks, one for each connection. Theoretically, these stacks could belong to different protocols, enabling a bridge to connect different types of networks. However, each type of network, such as Ethernet and Token Ring, has its own protocols at the data link layer. Translating data from the data link layer of an Ethernet to the data link layer of a Token Ring is difficult, but not impossible. Bridges, which operate at the data link layer, therefore, generally can join only networks of the same type. You see bridges employed most often in networks that are all Ethernet or all Token Ring. A few bridges have been marketed that can bridges networks that have different data link layers.
Figure 62 illustrates the protocol stack model for routing in terms of the OSI Reference Model. A different method of path determination can be employed using data found at the network layer. At that layer, networks are identified by logical network identifiers. This information can be used to build a picture of the network. This picture can be used to improve the efficiency of the paths that are chosen. Devices that forward data units based on network addresses are called routers. With TCP/IP, routing is a function of the internet layer. By convention, the network on which the data unit originates counts as one hop. Each time a data unit crosses a router, the hop count increases by one.
Figure 63 illustrates Hop-count routing.
• A-E-F (4 hops) • A-E-D-F (5 hops) • A-E-C-F (5 hops) • A-B-C-F (5 hops) By this method, A-E-F is the most efficient route. This assumes that all of the paths between the routers provide the same rate of service. A simple hop-count algorithm would be misleading if A-D and D-E were 1.5 Mbps lines while A-E was a 56 Kbps line. Apart from such extreme cases, however, hop-count routing is a definite improvement over no routing planning at all. Routing operates at the network layer. By the time data reach that layer, all evidence of the physical network has been shorn away. Both protocol stacks in the router can share a common network layer protocol. The network layer does not know or care if the network is Ethernet or Token Ring. Therefore, each stack can support different data link and physical layers. Consequently, routers posses a capability, fairly rare in bridges, to forward traffic between dissimilar types of networks. Owing to that capability, routers often are used to connect LAN’s to WAN’s. Building routers around the same protocol stack as are used on the end-nodes is possible. TCP/IP networks can use routers based on the same IP protocol employed at the workstation. However, it is not required that routers and end-nodes use the same routing protocol. Because network layers need not communicate with upper-layer protocols, different protocols may be used in routers than are used in the end-nodes. Commercial routers employ proprietary network layer protocols to perform routing. These custom protocols are among the keys to the improved routing performance provided by the bets routers. Circuit-based networks operate with high efficiency because the path is established once, when the circuit is established. Each switch maintains a table that records how data from different circuits should be switched. Switching is typically performed by lower-level protocols to enhance efficiency, and is associated most closely with the data link layer.
|
