Mạng và viễn thông P32

Chia sẻ: Hug Go Go | Ngày: | Loại File: PDF | Số trang:22

Thêm vào BST

Báo xấu

67
lượt xem 6
download

Download Vui lòng tải xuống để xem tài liệu đầy đủ

Network Trafic Control Traffic control is as much art as it is science, and the profits are gratifying. The right control mechanisms and routing algorithms in the right places will determine the overall performance and efficiency of our network. What is more, they will giveus simpler administration, better management and lower costs. This chapter describes a number of these admirable devices: first the simple methodscommonlyusedforoptimizingnetworkroutingundertypical‘normal-dayloading’ conditions; then we look at recent more powerful and complex techniques, together with some of the practical complications facing telecommunications today. ...

Chủ đề:

Bình luận(0) Đăng nhập để gửi bình luận!

Lưu

Nội dung Text: Mạng và viễn thông P32

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) 32 Network Trafic Control Traffic control is as much art as it is science, and the profits are gratifying. The right control mechanisms and routing algorithms in the right places will determine the overall performance and efficiency of our network. What is more, they will giveus simpler administration, better manage- ment and lower costs. This chapter describes a number of these admirable devices: first the simple methodscommonlyusedforoptimizingnetworkroutingundertypical‘normal-dayloading’ conditions; then we look at recent more powerful and complex techniques, together with some of the practical complications facing telecommunications today. Finally we note how several net- work operators have found ways interconnecting with the multiple carriers that have emerged of lately as a result of market deregulation. 3 . NETWORKS 21 Anymetropolitan,international,trunk,transitorcorporatenetwork, by virtue of its nature andsheer size has to include a considerable number of switches (exchanges). If there are many inter-connections between switches, individual these any call crossing the network will have plenty of alternative paths open to it. It is in fact the number of pathpermutations whichdeterminestherobustness of thenetwork to individual link failure and sets the level or grade of service provided (i.e. the prob- ability of successful call connection). It offers the customer what he most wants, an acceptablechance of makingasuccessfulcall or informationtransfereachtime. Particular attention must be paid to choosing the call-control mechanisms which are going to determine the routing of our customers’ individual calls, connections or mes- sages, bearing in mind that in an ideal network we aim overall to achieve a controlled flow of traffic throughout the network at reasonable cost and without the penalty of complicated administration. Various methods and preliminary calculations can now be considered in turn. 571
572 NETWORK TRAFFIC CONTROL 32.2 SIZING CIRCUIT-SWITCHED NETWORKS First of all, to find how many circuits a telecommunications network will need to meet traffic demand, a mathematical model is required to predict network performance, and it will comprise at least two parts 0 a statistical distribution to represent the number calls in progressat anygiven time of 0 a forecast of the overall volume of traffic Erlang’s model provides a statistical methodfor approximating telephone traffic, which is based on his measurements of practical telephone networks. A short reminder follows. Erlang devised a method of calculating the probability of any given number of calls being in progress at any instant in time. The probability density function used for the calculation is termed the Erlang distribution. The Erlang formula, a related and complex iterative mathematical formula, can be manipulated into a number of different forms. The most common form is used to calculate the required number of circuits (or circuit group size) to carry a given traffic volume between two exchanges, under theconstraint of having to meet a given grade of service. The traffic valueinput into this formula is the measurement of trafic intensity. The intensity of traffic on a route between any two exchanges in a network is equal to the average number of calls in progress, and is measured in Erlangs. In Chapter30 we discussed how, for planning purposes, it was normal measure the to route traffic intensity during the busiest hour of activity, or so-called route busy hour. Using this value and the Erlang formula, the number of circuits required for the route can be calculated according to a given target grade of service. The grade of service ( G O S ) of a telephone route between any two exchanges in a circuit-switched network is the fractional quantity of calls which cannot be completed due to network congestion. The lower the numerical value of GOS, the better the per- formance. A typical target value used in many trunk and international networks 1%, is or 0.01 (in other words 1% of calls cannot be completed due to network congestion, the other 99% can). The calling customer, however, probably only perceives end-to-end an grade of service, on a call-by-callbasis, of around 5 % (i.e. 5% lostcalls).This is because most connections comprise a number of links and exchanges, each of which is likely to be designed to inflict a 1% loss. The common method is to dimension circuit groups within a network using a target grade of service and the Erlang formula; so that Circuits Required = E(A, GOS) where E represents the appropriate form of the Erlang formula or function, with inputs A = route busy hour traffic and GOS = target grade of service. Alternatively, for packet or message-switched networks, we discussed an alternative formula, which sought to model the waiting time distribution packets, rather than the of proportion of lost calls; insteadof dimensioning in accordance with grade service, we of can dimension according to waiting time. The methodwas again covered Chapter 30. by
HIERARCHICAL NETWORK 573 32.3 HIERARCHICAL NETWORK The Erlang model is a good predictor of the traffic behaviour of most telecommunica- tions networks. A feature of telephone routes dimensioned using the Erlang method is that routes comprising a large number circuits are proportionately more of efficient (i.e. require proportionately fewer circuits per carried call) than smaller ones,when designed to the same grade of service, and are therefore more economic. This is apparent from the general shape of the graph of circuits against traffic-carrying capacity. Figure 32.1 illustrates the graph of 1% loss, or 1% grade of service. Note that the graph is almost 20 linear for circuit group with circuit numbers above circuits, each extra circuit adding approximately 0.85 Erlangs of traffic capacity. Thusby adding ten circuits to a group of Traffic capacity : l'/. g.0.s Number of circuits 30 l 110 20 10 6 0 10 20 30 Erlangs Figure 32.1- The inefficiency of small routes
574 NETWORK TRAFFIC CONTROL 20 circuits we make it suitable for carrying 20.5 Erlangs rather than 12 (an increase in capacity of 8.5 Erlangs). Notice that an isolated group of 10 circuits is only suitable for 4.5 Erlangs of offered traffic. Smaller circuit groups are thus far less efficient, as Figure 32.1 shows. Looking at the graph itis useful to imagine that the first six circuits carry notraffic at all, and that all subsequent circuits have a capacity of 0.85Erlangs each. It is almost as if there were a ‘penalty’ price of six circuits for having a route at all! The line representing this assumption is shown in Figure 32.1. The equation of this line provides a handy method of estimating the number of circuits required to carry a given offered traffic load, but it is only valid up to about A = Erlangs.Abovethis value, other estimating formulae can be used as follows N + (A/0.85) =6 for A < 75 N = 14 + (A/0.97) for 75 < A < 400 N=29+A 400 for
HIERARCHICAL NETWORK 575 International exchanges 3 ( 1 or 2 only) lncreaslng degree of inferconnection between exchanges t Trunk exchanges (relatively few) Localexchange ( many 1 Figure 32.3 Hierarchical network structure Figures 32.2 and 32.3. In such a structure, a small number of main exchanges are fully interconnected with one another, and various tiers of less important exchanges have progressively less directconnectionstootherexchanges.If no direct route is available from one given exchange to another, then the call is referred to an exchange in the next higher tier. By using such a structure we can reduce the number of circuits on long-haul routes. Under such a hierarchical scheme, routes are combined so that groups may be dimensioned (or sized) to carry multiple-traffic streams, and so reap the economic benefits of the larger scale. The hierarchial network technique works as shown in Figure 32.2. Let us assume that thecallers on each originating exchange A2, A3) in Figure 32.2 (Al, are generating calls equivalent to a traffic intensity of 25 Erlangs (i.e. an average of 25 simultaneous calls in progress) to each terminating exchange (Bl, B2, B3). Let us also assume that exchanges A l , A2, A3 are quiteclose together, as are exchanges B1, B2, B3, but that region A and region B are a long way apart. Without a hierarchical network structure, each A exchange would need a set of long- haul circuits to connect with each B exchange, as shown in Figure 32.2(a). This would give a total of nine routes (Al-B1, Al-B2, etc.). Using Erlang’s formula for 1% GOS, each of these routesor circuit groups wouldrequire 36 circuits,amassing a total requirement of 9 X 36 = 324 long haul circuits. Alternatively, if the traffic is concentrated through collecting exchanges T1 and T2 (called transit or tandem exchanges) withintheoriginating and terminating regions, only a single route is required. That route has to carry all 225 Erlangs, but only 247 long-haul circuits (again using Erlang’s formula) are needed. Figure 32.2(b) illustrates this alternative network structure. Of course, when comparing the cost of the the two structures we must not overlook the additional switches T1 and T2, as Comparison 1 below indicates. 32.3.1 Comparison 1: DirectversusHierarchicalStructure (Traffic between each A-B pair: 25 Erlangs) All direct routes 324 long circuits haul (9 X 36) (as Figure 32.2(a)) Hierarchical structure 247 long circuits haul (as Figure 32.2(b)) 2 X transit switches (Tl, T2); 225 Erlaags each 324 short haul circuits (A-TI and T2-B)
576 NETWORK TRAFFIC CONTROL So then, if 77 long-haul circuits (324-247) costs more than two 225 Erlang switches (Tl,T2) plus the short haul access circuits (A-T1 and T2-B), hierarchicalstructure (Figure 32.2(b)) is the more cost effective. In individual cases the particular circumstances will decide whether a direct or a hierarchical network structure is cheaper, though the general rule applies that in very- long-haulsituations(internationalnetworksforexample)thehierarchicalstructure usually wins. Hierarchical structure can also be cost-effective when point-to-point route traffic is very small (as in rural networks). To illustrate this point, in Comparison 2 the example of Figure 32.2 has been repeated with much lower traffic values (and with circuit numbers re-calculated, again using the Erlang method). 32.3.2 Comparison2:DirectversusHierachicalStructure: Low TrafficOnly (Traffic between each A-B pair: 3 Erlangs only) All direct routes 72 long circuits haul (9 X 8) (8 circuits required for each 3 Erlang route) Hierarchical structure 38 long circuits haul (for 27 Erlangs) 2 transit switches (Tl, T2); 27 Erlangs each 72 short haul circuits Note how in Comparison 2 the proportion of long-haul circuits saved as the result a of network hierarchy is much greater than it was in Comparison 1 (47% saving as opposed to 24%). This is because the inefficiency of very small routes is very marked, as we saw in Figure 32.1. There are therefore greater proportional savings to be made from com- bining very small routes. Hierarchicalstructure is common inmany of the world’s telephone and ISDN networks, in which a large number of local exchanges (or end offices) route their trunk traffic via a smaller number of trunk (or toll) exchanges. At the highest tier in the hierarchy there are probably only one or two international exchanges, each lower tier has a greater number of exchanges, but each with only a restricted degree of long-haul inter-connection. Figure 32.3 illustrates a typical national hierarchy. Anaddedadvantage enjoyed by hierarchicalnetworks lies intheircapacity for getting the most out of their circuits at times when the busy hours of various trans- mission routes do not coincide. For an example look back again at Figure 32.2(b), and we can see that if AI-B1 is busy in the morning and A3-B3 in the afternoon, then thanks to hierarchical structure the same circuits can be used for both traffic streams. Ontheotherhand,separate directcircuitgroups(asinFigure 32.2(a)) wouldbe inefficient, as one or other of the groups would . a l w a p be Idle. A disadvantage of hierarchical n e t w o r k - d e n compared with direct-circuited net- works, is their greater s u s c e p W t y to congestion under network overload. There are ~~~~~~ two causes. 0 Fewer overall circuits are available in hierarchical than in equivalent direct-circuited networks.
OVERFLOW OR 'AUTOMATIC ALTERNATIVE (AAR) ROUTING' 577 0 Congestion between only one pair of exchanges (e.g. AI-B1 of Figure 32.2(b)) will result in congestion on all otherroutes (e.g. A2-B3), because all calls have to compete for the same circuits (Tl-T2). This can rapidly lead to further congestion, as customers dial merrily on regardless. 32.4 OVERFLOW 'AUTOMATIC OR ALTERNATIVEROUTING' (AAR) A simple means of improving purely-hierarchial networks is to apply the technique of overflow, and for this we require the automatic alternative routing ( A A R ) mechanism. AAR is alsoknownas overflow or alternative routing, andmost exchangesare capable of it. It can be understood as a priority listing of the route choices leading to any given destination. Turning back to our example of Figure 32.2(b), let us assume that we introduce an additional link between exchanges A1 and B1. This might be a first choice, or high-usage ( H U ) route (as described below). In this case, an AAR table for traffic from A1 to B1 might be as shown in Figure 32.4. Cost benefits can be gained by judicious application of AAR (as in Figure 32.4) rather than using the simple hierarchical structure (as in Figure 32.2(b)), particularly if the route traffic between points A1 and B1 is large. The saving is achieved by making quite sure that the first choice route between A1 and B1 does not have enough circuits High-usage route I l 1 12 I I For traffic from A1 to B1 A A R t a b l e i s : A A Rt a b l e First hoice c Use o f t h e d i r e c t r o u t e to B1 Second choice O v e r f l o wt h et r a f f i c ( i f all f i r s t c h o i c e via T1 c i r c u i t s a r e busy) Figure 32.4 Automatic alternative routing (AAR)
578 NETWORK TRAFFIC CONTROL High-usage (HU) routes B1 A2 Final route 0 Tl T2 D - Figure 32.5 The benefit of high-usage working to carry the whole traffic. A remainder is then left which is forced to overflow via T1. Because of the way they are used, the two routes A1-B1 and AI-Tl-T2-B1 are called high-usage ( H U ) a n d j n a l routes, respectively. In fact A1-B1 is a primary high-usage route, as it is first choice; however, secondary, tertiary, and so on, high-usage routes (i.e. extra route choices between prirnaryHU a n d j n a l ) may also be used, and they have their place in the AAR table. To evaluate the savings of high-usage working, let us assume that we provide high- usage routesof 25 circuits between each pair exchanges in Figure 32.2(b) (i.e. nine of HU routes; A1-B1, Al-B2, etc). Then the network adopts the pattern shownFigure 32.5. in To compare the overall circuit requirement with earlier examples, must dimension we thenetwork tothe sameoverall 1% grade of service performance.However,the dimensioning of the overflow links(A-T1 and T2-B), as well as of the overflow orfinal route Tl-T2, presents a special problem for which Erlang formula has be modified. the to This is because overflow traffic does have arandom nature as the Erlang not dimensioning method requires. Consequently, in Comparison 3, the more complex Wilkinson-Rapp equivalent random method (explained later in the chapter) has been employed, and the network hasbeen dimensioned so that all customersreceive a gradeof service equivalent to or better than Yn. By this method, determine that 1 we 43jnal route circuitsare required + on Tl-T2. We thus need (9 X 25) 43 = 268 circuits in total. Comparison 3 below compares this value with the direct and the hierarchical network structures. 32.4.1 Comparison 3: DirectversusHierachicalversus HU Structure (Traffic between each A-B pair: 25 Erlangs) All direct routes: 324 long haul circuits Hierarchical structure: 247 long haul circuits 2 X transit switches (225 Erlangs each) 324 short haul circuits
RANDOMEQUIVALENT WILKINSON-RAPP 579 Overflow structure: 268 long haul circuits 2 X transit switches (32 Erlangs each) 99 short haul circuits Comparison 3 shows that it is possible (by choosing the optimum size of high-usage (HU) circuit groups) to reduce significantly the size of the transit switches required (between the hierarchical and overflow structures), without significantly increasing the long-haul circuit requirement. This clearly reduces the cost. However, our examplenotis very realistic. In practice, if any of the individual traffic streams are toosmall, the benefit ofproviding high-usage circuitsfor thecorrespondingdirectroutedisappears.In practice then, itis often useful to provide high-usage routes only thosetraffic streams for exceeding a given Erlang threshold value. In other words, ‘if more than X Erlangs of traffic exist between any two nodes, then a direct HU route is justified’. The value of X will depend upon the relative costs and of lineplant and exchanges equipment and on whether the exchanges are local, trunk, or international ones. A problem facing plannersusing overflow in practical networks is that the number of circuits required cannot always be cut to a minimum without adding another control mechanism such as trunk reservation, as explained later in this chapter. Furthermore, the high-usagestructure is notasadvantageousasthehierarchical when route busy hours of the various traffic streams do not coincide, because the cir- cuits are ‘less available’ for shared use, e.g. by onestream in the morning and by another in the afternoon. HU structures do, however, have a better overload performance, in that individual traffic streams are tosome extent protected from congestionon other routes (caused for example by very high seasonal or short term demand). 32.5 WILKINSON-RAPP EQUIVALENT RANDOMMETHOD In the previous section when we were attempting to dimension routes carrying overflow traffic we ran into the difficulty that because the distribution of traffic which is over- flowed from a high usage circuit group is not random, the Erlang formula may not be used directly. So, introduce the Wilkinson-Rapp equivalentrandom method instead. This is in fact the Erlang method slightly adapted. Imagine two traffic streams a and b which have high-usage routes of the A and B circuits available, respectively. These two traffic streams overflow amounts of traffic a’ and 6’ onto a common final route of N circuits, which is also the first choice for thejirst- oflered traffic stream c. Figure 32.6 illustrates this schematically. The problem is to find the value of N which will guarantee the appropriate grade service on all traffic streams. of + Circuit group N is subjected to overflow traffic a’ b’ and first-offered traffic c, and we dimension it by the Wilkinson-Rapp equivalentrandom method as follows. The method assumes that the Ncircuits will behave in the same as a set of Ncircuits part way way down a larger group of (NEQ+ N ) circuits when subjected to an imaginary single stream of equivalent random trafic, AEQ. Figure 32.7 illustrates the set of N circuits and the imaginary set NEQ. The problem is finding the values of NEQ and AEQ so that the traffic overflowed from the NEQ circuits exactly matches the characteristics of the real
580 NETWORK TRAFFIC CONTROL - H i gh usage Final routes route h r First-offered Overflowed traffic traffic(a') Q * A ccts D , First-offered Overflowed traffic B traffic (b') N b b ccts e ccts First-offeredtraffic C b Figure 32.6 Dimensioning final routes + + traffic a' b' c. The mean M and variance V characterize the traffica' b' c which + + we imagine to overflow from the N E Q circuits, and by choosing the values of AEQ and NEQ carefully we can equate the valuesexactlywith thecorrespondingmeanand variance values of the real traffic By knowing the valuesA,, and NEQ,the traffic lost by the N circuit group caneasily be determined. It is done by usingthe normalErlangformula, assuming thatan equivalent random traffic valueA,, is offered to a total number circuits N NEQ. of The + grade o service determined by the formula is the overall lost traffic ( L in Figure 32.7) f quoted as a proportion of the imaginary original traffic valueAEQ. 1 I G O S , , = LIA,o Imaginary Equivalent usage high random Real 'final' route traffic group M,V 1 N YL A EO c NE, ' / cct S * ccts ' Offered traffic matches Lost traffic i s assumed (a'+b'*c) to match the behaviour of real circuit group Figure 32.7 The Wilkinson-Rapp equivalent random method
DIMENSIONING ‘FINAL ROUTES’ 581 It is normal to dimension the group of N circuits by first calculating the permissible traffic loss in Erlangs (i.e. the amount of the real traffic which need not be carried, L in Figure 32.7), and then calculating this as a proportion of&Q. This is the imaginary or equivalent grade of service required when traffic A,, is offered to N + N E Q circuits, and + allows us to calculate the value of ( N NEQ) using the Erlang method as follows maximum permissible lost traffic = L L = GOS required X traffic on (a’ + b’ + c) imaginary GOS required on N + NEQ circuits = GOS,Q = L N+ N E Q = E(AEQ, GoSE,), where E represents the Erlang formula Having determined the value of N + NEQ, the real number of circuits required for the real traffic ( N ) is found by subtracting value NEQ. The values N E Q and AEQ are relatively straightforward to calculate, but the process requires a number of mathematical steps. An appendixat the end of the chapter shows how to calculate these values and gives an exampleof the wholeWilkinson-Rapp overflow route dimensioning method for those readers who are interested. TheWilkinson-Rappmethod is nowquiteold (1956) and in consequencemore complex methods have evolved whichseek to improve it.As with the alternatives to the basicErlangmethodit rests withtheuser to decidethemethodwhichsuits his circumstance the best. 32.6 DIMENSIONING ‘FINAL ROUTES’ It is normal practice to dimension final routes for only a 1% loss of the mean traffic offered to them. This ensures a 1% grade of service for any traffic which i s j r s t offered tothe final route.Thispractice was adopted in thepreeding section whenmixed overflow and first-offered traffic shared the same circuit group. In instances where there is no first ofSered traffic on the final route, then the circuit numbersmay be reduced,commensurate with an overall 1% loss oneach of the individual traffic streams. Thus if only 10% of traffic overflows from the high usage route (a typical value), only 10% of the original traffic is offered to the final route. The caller experiences a net1YOgrade-of-service evenif as muchas 10% of the traffic offered to thefinal route is lost. Thus the grade service of the final route need only be 10% in of this case! 32.7 TRUNK RESERVATION To some trunk reservation is a relatively new technique, for although it hasbeen around formorethan 30 years, only recently hastheadvent ofstoredprogramcontrol (SPC) exchangesmadeit more widely available. It providesamethod of priority
582 NETWORK TRAFFIC CONTROL ranking traffic streams, which can be particularly useful in conjunction with overflow network schemes, when the network planner may wish to give higher priority to Jirst- oflered trafic than to overflow trafic. As we saw in the example shown in Figure 32.6, final routes must be dimensioned large enough to ensure the chosen grade of service (usually 1Yo) any Jirst-offered for trafic streams.Theconsequence of this is an unavoidablyand unnecessarilygood grade-of-service for the overflowtraffic (far better than1%), and the penalty regrettably is the need to provide extra circuits over and above the minimum. Trunk reservation, however, gives a way out. Developing our example of Figure 32.2(b) a littlefurther to illustrate the principle of trunk reservation, let us now recognize that in practicetheroute traffic between individual A and B nodesis not identical in value and varies widely. Let us assume that exchanges A l , A2, B1, B2 and B3 are in large towns, and A3 is in a much smaller town and consequently generates and receives much less telephone traffic. It might well be that while A l , A2, B1,B2 and B3 justify direct interconnection by high-usage (HU) routes, overflowing via T1 and T2, the traffic originated at A3 might be insufficient to justify such direct HU groups. In this case the traffic passing over the circuit group from T1 and T2 is a mixture of 0 overflow (i.e. not all) traffic generated byA1 and A2 0 alltrafficgenerated by A3 Under such circumstance, it seems reasonable to take some precaution to ensure that the A3-originated traffic has some typeof priority for use of the Tl-T2 circuits so that the grade of service to A3 customers is on a par with the performance available to customers at AI and A2. Such a mechanism is provided by one form or another of trunk reservation. Numerous slightvariantsof trunk reservation exist. Inthesimplest,each traffic stream competing for agiven route is allocated a trunk reservation value (typical values range from 0 to 15). The value corresponds to a number of idle circuits. If at any given moment only this number of idle circuits are available within the group (the others being busy), then new calls from the particular traffic stream to which the trunk reserva- tion value applies, are rejected (i.e. are given network-busy tone and not completed). In other words, the trunk reservation value ‘disadvantages’ the traffic stream, to the benefit of other streams: the larger the value set, the larger the ‘disadvantage’. Each traffic stream competing for the circuit group may be allocated a different trunk reservation value (or disadvantage) butat least one stream should be given value 0 (top priority), otherwise some circuits will always be idle. By using trunk reservation we can ensure the that various traffic streams are completed according to a priority order during periods when the circuit group is heavily loaded (i.e. few circuits are idle). In this manner we can nearly equalize the grades of service of the trafffic streams, by disadvantaging overflow traffic streams which have had previous route options. When further circuits become free, fewer traffic streams are disadvantaged, so that more of the traffic streamsare allowed to complete calls. It is important to note that the algorithm does not take into account which particular circuits within the group are free, only the total number which are idle.
TRUNK RESERVATION 583 For our example illustrated in Figure 32.8 it would probably be appropriate to set values as shown in Figure 32.8. The mathematics needed to calculate trunk reservation values are exceptionally com- plex. However, it is found in practice that values of 3 or 4 establish a significant priority and can be used in a trial-and-error fashion. More accurate calculations are possible by the use of mathematical tables or computer programs, but as any one value of trunk reservation has much the same effect on all sizes of traffic streams, whether large or small, this only strengthens the case for trial-and-error use of values 3 and 4. Trunk reservation is highly effective inprovidingapriority list based on ‘dis- advantage’, and a value as low as 15 may virtually cutoff a given stream. As we shall see later in the chapter, routing techniques are moving from hierarchical (i.e. structured overflow) schemes towardsnon-hierarchical(i.e.dynamic) schemes. In consequence, new teletraffic modelling methods are necessary, capable of modelling end-to-end con- gestion rather than simple link-by-link analysis. Simple capacity flow models are used Trafficsource Trunk reservation value (applied at exchange 11, accordlng t o the source of t r a f f i c ) A3 0 (toppriority) A1 or A 2 3 o r L( t y p i c a lv a l u e ) Figure 32.8 Typical trunk reservation values
584 NETWORK TRAFFIC CONTROL for this analysis, but the technique of trunk reservation is no less relevant in protecting priority streams on particular links. In a dynamicroutingenvironment, trunk reservation valuesneed to be higher thanin simple overflow schemes (i.e. AAR), typically 10. 32.8 ‘CRANKBACK’ OR ‘AUTOMATIC RE-ROUTING’ (ARR) A more powerful form of overflow routing is that of automatic re-routing (ARR), also called crankback. This mechanism, which relies on sophisticatedswitch and signalling interaction, enables an alternative transit route to be chosen even if it is the second link of a previous transit route choice that is busy. Figure 32.9 shows how it works. Thismechanism is particularly useful inhighlyinterconnectednetworks;it gives much better grade-of-service where most of the available paths between two points are via transit exchanges, and it is made up of a multiple number of links. 1st choice C A B D 2nd choice Alternative routes A to B ARRfirst choice Via : C ( i ) i f A-C congestedthenARRvia A-D ( i i ) if A-C free and C-B free then forward c a l l to B otherwise if C-B b u s y go on to ARR second choice I ARR second choice V i a : D ( i f D-B b u s y then fail the call) Figure 32.9 Automatic re-routing (ARR)
BIDDING PROPORTIONATE 585 32.9 PROPORTIONATE BIDDING FACILITY (PBF) When there is more than one route toa given destination, it is sometimes an advantage to be able to control the exact proportion of traffic offered to each route. Proportionate bidding provides a mechanism for this purpose. Call attempts are ‘dealt’ in turn around all the available routes according to a set percentage regime (one for you, three for you, two for you, etc.). The ‘dealing’ is made by an exchange software program, which works in almost the same way that a croupier deals a pack of cards. Both calls and cards are subdivided into random fractional samples. Figure 32.10 provides an example. Useful applications of this algorithm might be as follows. Route balancing: suppose that exchange ‘A’ is a trunk exchange, and B and C are international exchanges with say 60% and 40%, respectively, of circuits to a given over- seas country. Traffic can be split by ‘A’ and routed toexchanges B and C in appropriate fractions (60/40), using PBF as a routing mechanism at exchange A. Multiple carrier environment: suppose this time that exchanges A, B, C, D are inter- national exchanges, A being in one country while B,C and D are owned by separate carriers (competing telecommunications companies) in another. Outgoing traffic from exchangeAcan be split, using PBF, in the same proportion as incoming traffic is received from carriersB, C and D. PBF in this case provides a fair proportionate return solution to the problems of connecting with multiple carriers. The method has already been adopted by a number of telephone network operators asa means of interconnect- ing with multiple telecommunications carriers operating in a destination country. 32.10 DYNAMIC ROUTING Another new technique is that of dynamic routing. A number of slightly different dynamic routing techniques have been developed, but all of them work on the same general premise: that the traffic pattern inalargenetwork is continuallychanging / l E- &*/. r o u t e d via B C./. r o u t e d via C . d./. r o u t e d via D &+c*d= 100 ‘ I . Figure 32.10 Proportionate bidding facility (PBF)
586 NETWORK TRAFFIC CONTROL time Pacific zone seaboard Eastern time zone U SA f: m New York Los Angeles Washington DC \ Figure 32.11 Dynamic routing / according to the time of day, the day of the week, and the season of the year. At any given time some routes will be busy while others are slack. If, then, one could employ some of the slack capacity to serve the busy routes, there would be an overall saving in network resources, even though some of the individual calls might have to be routed circuitously via exchanges in other timezones. Figure 32.11 shows an example of traffic being routed from New York to Washington via Los Angeles. In the Figure 32.1 1 example, as customers in Los Angeles are waking up and are starting to ring New York, the overall routing pattern needs to be adjusted dynam- ically, to prevent the New York/Washington traffic from causing congestion on the trans-USA links. At this time of day European and African exchanges will be down- loading as their business days end, and these switches could be used as alternative transit points for the New York/Washington traffic (instead of Los Angeles). Clever controlmechanismsareneeded to achieve optimum routing patterns, and to keep adapting the network so as to maintain the optimum. A number of such mechanisms havebeen patented. TheyincludeAmericanTelephone and Telegraph’s(AT&T’s) dynamic hierarchial non routing ( D N H R ) and British Telecom’s (BT’s) dynamic alternative routing ( D A R ) . 32.11 ROUTINGAND TRAFFIC CONTROL IN DATANETWORKS In contrast to circuit-switched networks, in particular the public telephone network, data networks (e.g. packet, frame, ATM, LAN, MAN and Znternet networks)have tended to use moresophisticatedautomaticroutingcontrol techniques.Typicalin telephonenetworks is thattheroutingtable ineachindividualexchangemustbe manually programmed (aswe saw in Figure 32.4). In data networks, more sophisticated routing protocols and algorithms (e.g. spanning tree protocol, a source routing protocol or some other protocol such as the TCP/IP method OSPF, open shortest path first) are generally used to select the route to the destination. The advantage is the simplicity of setting up maintaining and routing mechanisms in complicated networks,the disadvantage is the difficulty in measuring and predicting traffic flows. In the case of path-oriented, connection-oriented networks, all the packets, frames or cells take the same route through the network. Path orientation helps to ensure that the
ROUTING AND TRAFFIC CONTROL IN DTA NETWORKS 587 packets arrive in the same order in which they sent, thus minimizing thejob of any were necessary re-sequencing. The delays encountered by each of the packets is similar, thus giving approximately even performance of the network as perceived by the user. In addition, the path traffic volumes are relatively predictable and sometimes manually programmable using techniques similar to those discussed circuit-switched networks. for The alternatives to path-oriented networks are datagram networks. These are net- works in which individual packets constituting a given communication find their own individual way through thenetwork.Theadvantage of the datagram approach, as opposed to the path-oriented approach (both are illustratedinFigure 32.12) is the greater ability to spread traffic more evenly through the network according to instan- taneous data packet loading and availability of individual links. As not all the packets take the same route through a datagram network, it is more difficult for a third party to try to ‘tap’ the call, also making it more secure. Datagram networks are also more adept coping with individual link failures within network at the on a dynamic basis. Packets simply divert another route. When a link failure occurs via inapath-orientednetwork,thenetworkusuallytakesalittlelongertonoticethe failure, and then usually requires a little time to determine a new path. The virtual connection is not usually lost, but for maybe a few seconds (say around 10) it may not carry any further packets until the path is re-established over a new route. In connectionless networks, a virtual connection is not established. Instead, each of the packets or frames making up the message is simply despatched into the network in a manner similar to datagram switching. There is no confirmation of receipt of messages. The main advantage of connectionless networks is the effort saved in establishing con- nections and the high security resulting from the use of different paths and the absence of identifiable connections. Connectionless routing is used widely in the Internet. As is evident from the above, data networks have much more powerful automatic routing capabilities than has historically been the case in public telephone networks. This significantly reduces the manual effort required establish and administer routing to tables. In addition, it enables complex router networks (suchas the Internet itself) to be put together simply by plugging a large number of sub-networks operated by different organizations together, without the need for an organization planning the network and routing as a whole. So much for the advantages of automated routing algorithms and protocols; the disadvantage is the resulting chaos. The Internet, for example, has historically lacked a rigid structure and tight overall network planning authority, with the result that no-one Figure 32.12 Packet, frame or cell routing in data networks
588 NETWORK TRAFFIC CONTROL can accurately predict the path taken by an individual message traversing it. This leads to significant difficulties in designing an appropriate overall topology and ensuring adequate quality of network performance. Many customers complain of unacceptable response times when 'surfing' the Internet. To optimize the topologyof a data networkrequires a thorough understanding of the protocols and algorithms atwork in determining the routing of individual connections and packets or frames. Thus, for example, many different metrics and attributes may be used in the determination of the appropriate path, as we discussed in'Chapter 28. It is a difficult job to achieve optimum traffic flow through a datanetwork which uses complex routing protocols and algorithms. It requires very careful establishment within the network configuration database of the values of each of the metrics and attributes for each of the nodes and links, though this may be to some extent automated within the network management system. The easiest way to ensure success, as in telephone networks, is the use of a hierarchical network structure. Such a structure is rapidly appearing within the Internet, where only about four major nodes (all in the USA) switch most of the messages traversing the network. 32.12 NETWORK DESIGN A number of traffic-controlling mechanisms have been presented in this chapter. In practice, any one or any combination mechanisms may be used by network planners of in striving towards the perfect network; andalways the optimum topology makes some form of compromise between least cost, overall performance and robustness under overload. The constraints will arise from the control mechanisms actually available in any given case. None of the constraints and impositions are ever the same. All in all, network design is indeed a complex and imprecise art! Figure 32.13 provides a flowchart for network design, outlining the steps leading to the determination of a suitable topology. It starts by assuming that a network already exists and that the current point-to-point traffic demand between exchanges can be obtained by measurement. If this is not the case, current demand will have to be esti- mated. Alternatively, the scope for using an entirely new network configuration might be considered. Having measured the current traffic, the next step is to identify any problems of congestion or unreliable equipment. We must then estimate all the future requirements of the network, first by forecasting the future traffic demand (see Chapter 31), and then by reviewing any other reasons for undertaking network changes. The network mayneed to be changed to accommodate routes to new exchanges, or to connect to networks in previously unserved areas. Alternatively, its topology may need to change for economic or service reasons. Instead of being one made up of a large number of small outdated exchanges, it may become one comprising a small number of large, modern technology ones. Often these factors have the most impact on the evolv- ing network structure. When the future demands on the network are known, we review the efficiency and suit- ability of the current network. For its intended future purpose we determine as many dif- ferent options as possible for its evolution. At the simplest, an option may only require
NETWORK DESIGN 589 Measure the current point-to-point t r a f f i c demand Determine points of current network congestion or poor r e l i a b i l i t y Forecast future traffic demand and potential congestion Review any new network demands - new exchanges, modernization, other networks requiring interconnection etc Review the current network 1 efficiency and determine if new interconnections or new call control mechanisms are justified l Determine alternative network topologies, and compare the cost o f various options 1 Determine the best topology and calculate circuit numbers, together w i t h any new routes and exchanges t h a t are needed 1 Check t h a t t h e r e l i a b i l i t y and congestion objectives are met Figure 32.13 Network design
590 NETWORK TRAFFIC CONTROL adjustment to the capacity the exchanges or of the transmission links interconnecting of them. Morecomplex options might demand theuse of new call control techniques (such as ARR, PBF ordynamic routing). However, the deployment such techniquesis likely of to incur considerable cost and needs to be weighed against the benefits, both in overall network performance and in overall network cost. Once the choice has been made, the final circuit numbers can be determined and a plan worked out for the new circuits, routes and exchanges that will be needed. Then it is up to the implementors. Depending on thenature of thenetworkand itscustomers,theplan will need periodic, if not frequent, review and readjustment. Nonetheless, the more comprehen- sive and forward looking is the plan, the greater is the likelihood of smooth network administration and performance. Contrariwise, insufficient forward forecasting or lack of anticipation of future needs can lead to short-term ‘firefighting’, with a need to adjust continually the structureof the network. Under such circumstances few networks reach their optimum of performance and economic efficiency. 32.13 APPENDIX: THE WILKINSON-RAPP ROUTE DIMENSIONINGMETHOD In the main part of the text we briefly describedtheWilkinson-Rappmethod for dimensioning overflow circuit groups, but althoughwe described the method we did not demonstrate how to calculate the values AEQ and NEQ. For those interested readers we now cover the necessary calculations and give an example of the use of the method. The explanation refers back to the diagrams of Figures 32.6 and 32,7 and describes the method in mathematical notation. In Figure 32.6 the traffic offered to the N circuit group is a’ + h’ overflow traffic c first-offered traffic Thequestion is ‘how many circuits ( N ) are requiredin orderthat a 1% GOS is achieved?’ The value can be calculated as below. First, the separate streams within the offered traffic are characterized in mathematical terms by the statisticalmeans and variances of their traffic intensities,which are calculated using the normal Erlang formula. Provided that the streams are statistically independent of oneanother,thenthemeanand variance of thecombinedstream a’ + b’ + c can be calculated by simple summation. These values in turn allow A,, and N E , to be worked out. The method is then as explained in the main text of the chapter.Refer to Figure 32.7. The actualtraffic which may be lost,L , is calculated as 1 % X (a’ + h’ + c) and from this the equivalent grade of service GOS,, can be worked out. + Finally, the total size of the equivalent circuit group N NEQ is obtained from the normal Erlang formula and the value of N deduced by subtraction of value NEQ. There now follows a short summary in mathematical notation.