Tài liệu Mạng và viễn thông P18 ppt

PART 3 MODERN DATA NETWORKS Networks and Telecommunications: Design and Operation, Second Edition. Martin P. Clark Copyright © 1991, 1997 John Wiley & Sons Ltd ISBNs: 0-471-97346-7 (Hardback); 0-470-84158-3 (Electronic) Packet Switching Packet switching emerged in the 1970s as an efficient means of data conveyance. It overcame the inability of circuit-switched (telephone) networks to provide efficiently for variable bandwidth connections for bursty-type usage as required between computers, terminals and storage devices. In this chapter we discuss the basics of packet switching and ITU-T’s X.25 recommendation, nowadays the worldwide technical standard interface to packet-switched networks. We then also go on to discuss the IBM company’s SNA (systems network architecture), a proprietary form of packet switching, important because of its dominant role in IBM computer networks. 18.1 PACKET SWITCHING BASICS Packet switching is so-called because the user’s overall message is broken up into a number of smaller packets, each of which is sent separately. We illustrated the concept in Figure 1.10 of Chapter 1. Each packet of data is labelled to identify its intended destination, and protocol control information (PCZ) is added, as we saw in Chapter 9, before it is sent. The receiving end re-assembles the packets in their proper order, with the aid of sequence numbers and the other PC1 fields. Each packet is carried across the network in a store-and-forward fashion, taking the most efficient route available at the time. Packet switching is a form of statistical multiplexing, as we discovered in Chapter 9. Figure 18.1 illustrates how a link within a packet switching network is used to carry the jumbled-up packets of various different messages and the use of the information carried in the packet header to sort arriving packets at the destination end into the separate logical channels, virtual circuits ( VCs) or virtual calls (VCs). Transmission capacity between pairs of nodes in a packet-switched network is generally not split up into rigidly separate physical channels, each of a fixed bandwidth. Instead, the entire available bandwidth between two nodal points (switches) in the network is bundled together as a single high bitrate pipe, and all packets to be sent between the two endpoints of the link share the same pipe (Figure 18.1). In this way, the entire bandwidth (i.e. full bitspeed) can be used momentarily by any of the logical channels sharing the connection. This means that individual packets are transported more quickly and bursts of transmission can be accommodated. 341 Networks and Telecommunications: Design and Operation, Second Edition. Martin P. Clark Copyright © 1991, 1997 John Wiley & Sons Ltd ISBNs: 0-471-97346-7 (Hardback); 0-470-84158-3 (Electronic) 342 PACKET SWITCHING n packet may mm switch U Figure 18.1 The statistical multiplexing principle of packet switching A problem arises when more than one or all logical channels try to send packets at once. This is accommodated by buffers at sending and receiving ends of the connection as shown in Figure 18.2. These delay some of the simultaneous packets for an instant until the line becomes free. By use of buffers as shown in Figure 18.2, it is possible to run the transmission link at very close to 100% utilization. This is achieved by sharing the capacity between a number of end devices (each with a logical channel). The statistical average of the total bitrate of all the logical channels must be slightly lower than the line bitrate so that all packets may be carried, but at any individual point in time the buffers may be accumulating packets or emptying their contents to the line. Packet switching is able to carry logical channels of almost any average bitrate. Thus a 128 kbit/s trunk between two packet switches might carry 6 logical channels of mixed and varying bitrates 5.6 kbit/s, 11.4 kbit/s, 12.3 kbit/s, 22.1 kbit/s, 28.7 kbit/s, 43.0 kbit/s and still have capacity to spare. This compares with the two channels which a telephone network would be able to carry using the same trunk capacity. (The excess capacity of the telephone channels simply has to be wasted, and the other four channels cannot be carried.) packet switch buff er Figure 18.2 The use of a buffer to accommodate simultaneous sending of packets by different logical channels TRANSMISSION DELAY IN PACKET-SWITCHED NETWORKS 343 18.2 TRANSMISSION DELAY IN PACKET-SWITCHED NETWORKS When using the trunks in a packet-switched network at very close to full utilization, very large buffers are required for each of the logical channels, to smooth out the bursts from individual channels into a smooth output for carriage by the line. (This is rather like having a very large water reservoir, collecting water during showers of rain, and varying in water depth, but always capable of outputting a constant volume of water for munic- ipal use (Figure 18.3). The water reservoir is analagous to the data buffers, the showers of rain to the bursts of data information, and the constant output to the information carried by the line.) We can make sure that the packets accumulated in the buffer are despatched on a jirst-in-jirst out (FIFO) basis to fairly share out the queueing delays which result, but it is critical to ensure that the queueing delay does not become unacceptably long. The chance of a very long delay is much greater when close to 100% utilization of the line is expected. (Imagine waiting in line for a bus, all of the seats of which had to be full before it pulled away; either the bus doesn’t come very often, or there is a very long queue to ensure that all the seats can be filled). A certain amount of queueing delay caused by buffering is not noticeable to computer users (a $ second is a very long queueing delay in packet switching network terms). Even if a typed character did not appear on the computer screen until a f second after hitting the keyboard, the user is unlikely to notice. A variation in the delay (sometimes a $ second, and sometimes no delay) is also unimportant. (The fact that some characters appear on the screen more quickly than the f second maximum delay will not be noticed.) On the other hand, once the average delay becomes much longer, then computer work may become frustrating, so that much longer queueing delays are unacceptable. There is an entire statistical science used to estimate queueing delays. The most important formula is the Erlang call-waiting formula, which we will discuss in Chapter 30. In simple terms, however, the unacceptability of long queueing delays means that the 344 PACKET SWITCHING trunks in packet-switching networks may not be utilized at 100%. Typical acceptable maximum utilization is around 50%. Despite this fact, they are still more efficient than circuit-switched networks to carry data tra@c (i.e. information between computers). 18.3 ROUTING IN PACKET-SWITCHED NETWORKS Packets are routed across the individual paths within the network according to the prevailing traffic conditions, the link error reliability, and the shortest path to the desti- nation, according to one of two main techniques, so-called path-oriented routing and dutagram routing. The routes chosen are controlled by the (layer 3) software of the packet switch, together with routing information pre-set by the network operator. Path-oriented routing is nowadays the most common technique. In path-oriented routing, a fixed path is chosen for a given logical channel (i.e. virtual circuit, VC) at the time of call set up. The path itself is chosen based on the current loading of the network and the available topology. In the case of the virtual call (i.e. switched virtual con- nection) service, the required destination of the path is indicated using an X.121 address carried by a layer 3 packet, called the call set-up packet. This packet has the equivalent function to the dialled digits of a telephone number in setting up a telephone call. Should any link in the path become unavailable during the course of the call (say because of a transmission failure), then an altenative path is sought, without breaking the connection. The packets are stored for a short while, while the new path is found and then sent over this path. (Figure 18.4). The advantage of path-oriented routing is that the packets pertaining to a given logical connection all take the same path, all suffer about the same amount of queue- ing delay in the buffers and arrive pretty much in the same order as they were sent (allowing for any lost on the way). This makes the job of re-sequencing the packets at the receiving end much easier, as well as the job of directing the packets through the network. It also leads to more predictable delay performance for the end user or computer application. The packet switching network components can therefore be relatively simple and cheap. The disadvantage of path-oriented routing is the inability of the network on a more immediate basis to employ alternative routing to better utilize the network as a whole (A31ppq-W back-up C-path c1 p1 (c31 4q m p1 packet 4 packet * switch fcl normal A-connection-path ," normal C-path (currently failed) A' Figure 18.4 Circumventing a transmission link failure using path-oriented routing ROUTING IN PACKET-SWITCHED NETWORKS 345 m L U U) 2 346 PACKET SWITCHING during periods of sudden surge in demand resulting from simultaneous packet bursts by many logical channels sharing the same path. The second type of routing, datagram routing, allows for more dynamic routing of individual packets (Figure 18.5), and thus has the potential for better overall network efficiency. The technique, however, requires more sophisticated equipment, and powerful switch processors capable of determining routes for individual packets. Packet switching gives good end-to-end reliability, with well-designed switches and networks it is possible to bypass network failures (even during the progress of a call). Packet switching is also efficient in its use of network links and resources, sharing them between a number of calls, thereby increasing their utilization. 18.4 ITU-T RECOMMENDATION X.25 Most packet-switched networks use the protocol standards set by ITU-T’s recommen- dation X.25. This sets out the manner in which a data terminal equipment (DTE) should interact with a data circuit terminating equipment (DCE), forming the interface to a packet-switched network. The relationship is shown in Figure 18.6. The X.25 recommendation defines the protocols between DTE (e.g. personal computer or computer terminal controller (e.g. IBM 3174)) and DCE (i.e. the connection point to a wide area network, WAN) corresponding to OS1 layers 1, 2 and 3 (Figure 18.7) which we learned about in Chapter 9). The physical connection may either be X.21 (digital leaseline) or X.21 bis (V.24/V.28 modem in conjunction with an analogue leaseline: Chapter 9). Alternatively, the X.31 recommendation (Chapter 10) specifies how the physical connection (DTE/DCE) may be achieved via an ISDN (integrated digital services network). Finally, recommenda- tion X.32 specifies the use of a dial-up connection for a packet mode connection via the telephone or ISDN network to an X.25 packet exchange. The X.25 recommendation itself defines the OS1 Iayer 2 and layer 3 protocols. These are called the link access procedures (LAPB and LAP) and the packet level interface. The link access procedure assures the correct carriage of data across the link connecting DTE DTE DTE DCE DCE I I I I l *X25 U IcX25-W Packet switched network Figure 18.6 The X.25 interface to packet switched networks ITU-T RECOMMENDATION X.25 347 layer 3 protocol (packet layer) layer 2 protocol (link layer) layer 1 protocol (physical layer) (NETWORK) E DTE X.25 packet level interface * - X.25 LAPB (link access procedure) X.21, X.Slbis, X.31 or X.32 DCE Figure 18.7 OS1 layered model representation of ITU-T recommendation X.25 to DCE and for multiplexing of logical channels; the packet level interface meanwhile guarantees the end-to-end carriage of information across the network as a whole. The LAPB (link access procedure balanced) protocol provides for the IS0 balanced class of procedure and also allows for use of multiple physical circuits making up a single logical link. LAP is an older and simpler procedure only suitable for single physical circuits without balanced operation. The link access procedures use the principles and terminology of high-level data link control (HDLC) as defined by ISO. This procedure ensures the correct and error-free transmission of data information across the link from DTE to DCE. It does not, however, enable the DCE (in the form of a data switching exchange, DSE) to determine where the information should be forwarded to within the network or ensure its correct and error-free arrival at the distant side of the packet network. This is the job of the OS1 layer 3 protocol, the X.25 packet level interface. During the set-up of a switched virtual circuit (SVC, also called a virtual call (VC)) it is a level 3 call set up packet which delivers the DSE the data network address of the remote DTE. Level 3 packets confirm the set-up of the connection to the initiating DTE and then pass end to end through the network, allowing user data to be carried between the DTEs. A packet of data carried by the X.25 protocol may be anything between three and about 4100octets (bytes: 8 bits). Up to 4096 alphanumeric characters of user information may be carried in a single packet. In slang usage, many people refer to ‘X.25 networks’. In general they mean packet switching networks to which X.25 compliant DTEs may be connected, for recommenda- tion X.25 describes only the UNI (user-network interface, see Chapter 7). The X.25 protocol allows DTEs made by any manufacturer to communicate across the network. You should not be tempted into believing that the protocol used between the various packet data switching exchanges (DSEs) within the network is also X.25. Generally, packet-switched data networks are built from a number of individual exchanges, but all of them provided by the same manufacturer. The protocol used for the carriage of the data between the exchanges is normally an X.25-like, but enhanced ‘proprietary’ protocol. Examples include those used by Northern Telecom (Nortel), Telenet, BBS, Tymnet and France’s Transpac. Proprietary trunk protocols typically allow the carriage of sophisticated network management and charging information back to the network control centre. In addition, 348 PACKET SWITCHING they may allow for dynamic adjustment of the traffic paths taken through the network, so giving better overall network performance during heavy traffic loading. Where separate packet-switched sub-networks (provided by different manufac- turers) need to be interconnected, the X.75 (NNI) protocol is used. We discuss this later in the chapter. 18.5 THE TECHNICAL DETAILS OF X.25 X.25 was one of the first data protocols to be well defined and standardized. As such it has formed the basis on which later data transport protocols have been developed. Understanding the principles in detail will give the reader a very good understanding of all other data switching protocols, all of which use similar principles. There thus follows a very detailed description. 18.6 X.25 LINK ACCESS PROCEDURE (LAP AND LAPB) The link access procedure can be performed either in the basic mode (B = 0, called LAP) or in the more advanced balanced mode (B = 1, called LAPB). Nowadays the LAPB mode is more common. There are two forms of LAPB; the basic form is called LAPB modulo 8, the extended form is called LAPB modulo 128. Only the modulo 8 form is universally available. The difference between the two forms is only the maximum value of the sequence number given to consecutive packets before resetting to value ‘0’. LAPB allows for data frames to be carried across a physical layer connection between a DTE and a DCE. The frame is structured in the manner shown in Figure 18.8. The jag is a delimiter between frames. The address (perhaps confusingly named, as we discovered in Chapter 9) is a means of indicating whether the frame is a command or a response frame, and whether control is with the DTE or DCE. It is coded as shown in Table 18.1. Flag Flag Frame Check Information Control Address 01111110 (of next frame) Sequence (N bits) (8 bits) (8 bits) (16 bits) Figure 18.8 X.25 LAPB modulo 8 frame Table 18.1 X.25 LAPB address field coding Single link procedure Multiple link procedure command from DCE to DTE address A - 1 l000000 address C - 1 11 10000 response of DTE to DCE address A - 1 l000000 address C - 11 110000 command from DTE to DCE address B - 10000000 address D - 11 l00000 response of DCE to DTE address B - 10000000 address D - 11 l00000 X.25 LINK ACCESS PROCEDURE (LAP AND LAPB) 349 Table 18.2 X.25 LAPB modulo 8 control field command and response coding Format Command Response bit 1 bit 2 bit 3 bit 4 bit 5 bit 6 bit 7 bit 8 ~ ~ ~ ~ ~ ~~~~~~~~~~~~ Information I (information) transfer Supervisory RR (receive ready) RNR (receive not ready) REJ (reject) Unnumbered SABM (set asynchronous balanced mode) DISC (disconnect) DM (disconnect mode) UA (unnumbered acknowledgement) FRMR (frame reject) 0 P 1100P010 llllFllO llOOFllO lllOFOOl The controlfieldcontains either a command or a response, and sequence numbers where applicable as a reference when acknowledging receipt of a previous frame. There are three types of control field format, corresponding to the numbered information trans- fer of I-frames (I format), numbered supervisory functions (S format) and unnumbered control functions (U format). The control field is coded as detailed in Table 18.2. The frame check sequence (FCS) field is a string of bits which help to determine at the receiving end whether the data in the frame has in any way been corrupted. It is a 16 bit field, created using the properties of a cyclic code, hence the term cyclic redundancy check. The exact 16 bit sequence sent is the ones complement of the sum (in binary) of the following two parts. (1) The remainder of xk(xI5 + xI4 + xI3 + XI* + xl1 + x10 + x9 + x8 + x7 + x6 +X5 + X4 + x3 + X2 + XI + X + 1) divided (in binary) by x16 + xL2 + x5 + 1 where k is the number of bits in the frame excluding flag and FCS. (2) The remainder of the product of the frame content (excluding flag and FCS) and xI6, when divided by X'6 + X12 + X5 + 1 There are six configurable parameters when setting up LAPB connections. These, their meaning and typical value settings are given in Table 18.3.