For network dimensioning the models describing the traffic mixes should be as realistic as possible. The real patterns of everyday traffic are complicated to model analytically. Many user related and system parameters have to be defined in order to make the traffic model realistic. This paper presents a simulation environment for generating multislot traffic by taking into account the behavior of A-subscriber and B-subscriber. Their behavior has been defined by dialing delays, answering delay, absence probability, repetition intervals, repetition probabilities etc. Different trunk reservation schemes have been used in the allocation of bandwidth for new multislot call. The feedback effect of system blocking and user behavior were observed at the overload condition with different repetition parameters. The traffic generation model consists of an option to change the traffic intensities with respect to time. For example, the influence of a sudden increase in traffic due to "T.V. rush" has been studied. With the detailed measurement part one can observe not only the steady state but also the transient behavior. The developed software can be used as an educational tool for studying the traffic characteristic of ISDN calls with different user behavior parameters.

1. Introduction
2. Simulation Environment
3. Repeated Attempts
4. Traffic Mixing
5. Concluding Remarks
6. References

1. INTRODUCTION

Traffic which is a consequence of inter-human behavior is correlated to the customs of subscribers and the society, where it is generated. The habits of receivers also influence the traffic pattern. With the advent of ISDN in digital exchanges a wide variety of calls such as voice, data and video with heterogeneous types of traffic is required to be served by common shared resources of the exchange. This may have undesirable effects on the performance of individual services.

User’s reaction to the progress of call and repetition of failed calls is another important issue. Theoretical studies often assume that call arrival process obeys a Poisson distribution and the blocked calls leave the system finally. In the case of real traffic the subscriber who receives the busy tone will often repeat the call after a more or less brief interval. There will be a large number of abandonments and rejections due to reasons such as B-busy, B-late answering etc., and resulting reattempts will markedly modify the nature of the offered traffic [2].

In addition to user dependent durations, setup and release times of certain common equipment should be taken into account. For example, when determining the total traffic load of a system, common equipment setup times become more significant in the presence of voice traffic overloads. Some typical values of parameters related to user behavior and call failures are presented in Table 1 [2].

The focus of this paper is on creating the simulation environment for ISDN switching system with user dependent traffic. The developed software can be used as an educational tool to study the behavior of multi-service traffic at transient states and at steady state.

The block scheme of simulation environment consisting of switching system, traffic source and sink, and monitoring module is shown in Figure 1. The implemented simple switching system for handling the calls consists of dialing queue, call setup queue and user defined number of dial digit receivers, call processors and traffic channels fully available from and to the subscribers. The links work as a loss system whereas dial receivers and processors work as waiting systems. The receivers and processors are used for a short constant processing time given by the user in the order of 10 sec and 50 ms ... 500 ms respectively. Maximum queue size can be specified by the user. Each accepted or processed call holds a link for two distinct time intervals, a call setup time and communication time. Dialing time is assumed to be normally distributed with mean value of 7 sec and variance 2 sec.

A large number of parameters have been specified in order to take into account the behavior of the users and the performance of the system as presented in Table 2. The model consists of a fixed user defined number of subscribers generating traffic with separate ranges of subscriber numbers for different service groups. The subscriber groups are divided according to the bandwidth required for each service. Maximum number of possible services is 12 currently. Subscribers in the same group can also be divided according to their mean arrival rates and mean holding times to generate calls e.g from residential area and business area. The traffic source consists of user behavior parameters, time varying traffic parameters and subscriber categories for multi-service calls. All the subscriber related reaction times are supposed to be exponentially distributed.

Time varying traffic is generated by allowing the simulation program to change the traffic characteristics e.g. arrival rate, holding time and bandwidth at flexible time intervals. The traffic offered by group i is modeled by fresh arrival rate l _i, mean holding time tm_i, repetition probabilities, failure probabilities and user related time parameters. For example services 1, 2 and 3 are characterized by offered traffic A_i = l _i.tm_i in Erlangs. To account for the different bandwidth demands, the service i traffic load is defined as A_{ib i}, where b _i = basic bandwidth unit (Number of time slots). The total fresh offered traffic load generated is related to the traffic load of all services according to following relation:

A = A_{1b 1}+A_{2b 2}+A_{3b 3}                        (1)

There are many reasons for the call to fail which makes user to reattempt causing total offered traffic to differ from the fresh traffic. These reasons are characterised by different probabilities and time parameters such as B-busy, A-waiting time etc., and in the current software they are valid for all the traffic categories. The detailed flow diagram describing the chain of events when generating a call is shown in Figure 2.

A separate monitoring module has been implemented in order to study detailed phenomena of traffic process. The on line monitoring is made possible by updating the monitored parameters at every event. The software includes modules to collect statistics about offered and carried traffic intensities and distributions, arrival rates, holding times, call blocking, and time congestion per service group. Scanning method with user defined scanning interval is used to measure traffic intensities. Occurrence and duration of individual congestion states can also be recorded for each service category.

3. REPEATED ATTEMPTS

Repeated attempts distort the original flow of calls in a predictable manner and create an overload in the system. The origin of this effect may be due to fault of equipment, or a sudden peak traffic called "T.V. rush".

Figure 3 shows a traffic pattern constructed by the simulation tool following quite accurately the original traffic measurement results of a "T.V. rush" on a 90-line circuit group [13]. This pattern was created by letting traffic intensities vary during a period of 15 minutes and using a multiple set of control parameters. For comparison the same measurements were performed without repetitions. In the case of reattempts traffic increases faster than without reattempts because the traffic is forced to grow with maximal speed. During the intervals t4 and t5, when T.V. rush is over, additional delayed traffic caused by reattempts can be seen also in the case of small load. The differences of the runs during periods t1 and t2 are also due to the fact that links in the model are used for both control and communication purposes and there is a significant amount of holding times such as dialing time and A-waiting time which load the system even in the case of unsuccessful call. Thus the reattempts due to failed calls increase the carried traffic load.

Figure 4 presents the effect of repeated attempts in the case of multi-service calls served by a fully shared group of 30 channels. The offered wide band traffic mean is constant, A2 =1 Erlang (bandwidth is 5 time slots), and the offered narrow band (BW = 1 slot) traffic A1 varies from 10 to 40 Erlangs. The holding times for narrow band and wide band calls are 200 s and 1000 s respectively. Repeating probabilities and mean intervals for consecutive trials are 50%, 40%, 30% and 30 s, 90 s, 240 s. Here the repetition of call is consequence of the system blocking. Other failure reasons such as B-absent, B-late answering etc. have been ignored in this case. Since the repetition intervals are very short compared to the holding times, the repeated calls will find the system congested with high probability. Because of no protection method is applied, the wide band traffic experiences much higher blocking than the narrow band traffic.

Multi-service traffic was offered to a group of 30 channels with trunk reservation scheme. Figure 5 shows instantaneous occupation states of carried wide band and narrow band traffic components during approximately 3 hours observation time obtained by scanning at 100 sec intervals. The mixture consists of two traffic flows: l 1=0.175 calls/sec, tm1=200 sec, BW1=1, TR1=4, and l 2=0.001 calls/sec, tm1=1000 sec, BW2=5, TR2=0. According to traffic measurements time congestions for both service components are equal (39.4%), but call blockings differ slightly (B1 = 39% and B2 = 32.4%). Due to the trunk reservation the congestion states of the narrow band occur frequently at occupation state x = 26. Also, certain congestion states have been observed at x = 21 (26-5) and x = 6 (21-5) since the channels are occupied by wide band service. Because the wide band traffic appears very seldom, the trunk reservation scheme decreases the traffic utilization though it balances the blockings. As seen in the Figure 5, four last channels are free most of the time and they are reserved for wide band service. Thus the available channels are not utilized efficiently.

Influence of trunk reservation parameter on call blocking and channel utilization is presented in Figure 6. Offered traffics are A1=15 Erlangs (BW1 = 1), and A 2 =1.5 Erlangs (BW2 = 10). When increasing the protection of wide band traffic the overall utilization decreases, but call congestions and time congestions are balanced for both services in the case of full protection.

The similar phenomenon can be observed in Figure 7. It shows that the carried traffic in the case of trunk reservation is always less than in the case of complete sharing (TR=0). The gap is more pronounced when the proportion of wide band traffic is low, since the reservation scheme reserves the trunks for wide band service. Increasing the proportion of service 2 traffic mix in the case of complete sharing scheme does not effect much on the blocking when the traffic is low, but however, at high traffic intensity there is an optimum mixing point to get the best overall GOS, for A=45 Erlangs it is at the point of 70% service 2 mix.

Figure 9 presents carried traffic distributions for a mixture of wide band and narrow band services. It can be seen that the overall carried traffic is widely distributed with peakedness z = 2.18, although the congestion has been about 10%. The distribution of traffic for narrow band service is smooth (z = 0.938), but the wide band traffic covers the whole possible range thus having high peakedness (z = 2.76). Blocking values for narrow band and wide band components have been about 6%, and 30% respectively.

In digital exchanges a wide variety of calls such as voice, data and video are required to be served by common shared resources of the exchange with specified quality of service performance. The models describing multi-slot traffic mixes used for system dimensioning should be as realistic as possible. In the case of several traffic components theoretical modeling becomes extremely complicated.

In this paper a simulation environment for circuit-switched ISDN communication system with user dependent traffic has been introduced. The behaviors of A-subscribers and B-subscribers have been defined by dialing delays, answering delay, absence probability, repetition intervals, repetition probabilities etc. Trunk reservation schemes have been used in the allocation of bandwidth for new multislot calls connected through a simple switching system. The traffic generation model consists of an option of time dependent traffic. With the detailed measurement part the performance of the communication system can be observed not only in the steady state but also during various transient situations. The developed software can be used as an educational tool to study ISDN traffic characteristics and for example the introduction of some new service.

6. REFERENCES

1. REDMILL, F.J. & VALDER, A.R., SPC Digital Telephone Exchange, IEE Telecommunications Series 2 ,1990.
2. HEBUTERNE, G., Traffic Flow in Switching System, Artech House, 1987.
3. FILIPAK, J. & CHEMOUIL, P., Modeling and Prediction of Traffic Fluctuation in Telephone Networks, IEEE Trans. on Communications, Vol 35 No. 9, 1987.
4. PIDD, M., Computer Modeling for Discrete Simulation, John Wiley & Sons, 1989.
5. GRABOWSKI, K.H., & HAGENHAUL, L., Traffic Model for ISDN with Integrated Packet Switching, ITC 12, Torino, June 1988.
6. HABIB, I.W. & SAADAWI, T.N., Multimedia Traffic Characteristics in Broadband Networks, IEEE Communications magazine, July 1992.
7. LINDBERGER, K., Blocking for Multislot Heterogeneous Traffic Streams Offered to Trunk Group with Reservation, ITC 12, June1988.
8. ODA, T. & WATANABE, Y., Optimal Trunk Reservation for Group with Multislot Traffic Streams, IEEE Transactions on Communications Vol. 38, No. 7, 1990.
9. TAKAGI, T. & SAKITA, Y., Analysis of Loss Probability Equalized by Trunk Reservation for Mixture of Several Bandwidth Traffic, ITC 12, June 1988.
10. SCHWARTZ, M., Telecommunication Networks, Protocols, Modeling and Analysis, Addison-Wesley, 1988.
11. YAMADA, H & SUZUKI, Y., Network Configuration for New Services Traffic. ITC-12, June 1988.
12. LINDBERG, P., NIVERT, K., & SAGERHOLM, B., Trunk Reservation and Grade of Service Issues in Circuit Switched Integrated Networks, ITC 12, June 1988.
13. RAHKO, K., HERTZBERG, S., Traffic Measurement-TETRAPRO Special Course, Helsinki University of Technology ,1988
14. FICHE, G., CORGUILLE, D. AND PALUD, C., Traffic Model for an ISDN Switching System, ITC 12, June 1988.
15. HOFSTETTER, H. AND WEBER, D., Traffic Models for Large ISDN - PABX’s, North Holland, Computer Networks and ISDN Systems, 1985 .
16. KLEINEWILLINGHOFER - KOPP, R. AND WOLLNER, E., Comparision of Access Control Strategies For ISDN-Traffic On Common Trunk Groups, ITC 12, 1989.