Performance Analysis of Integrated Voice and Data Systems Considering Different Service Distributions

Transcription

1 Performance Analysis of Integrated Voice and Data Systems Considering Different Service Distributions Eser Gemikonakli University of Kyrenia, Kyrenia, Mersin 10, Turkey Abstract In this study, the aim was to evaluate the effect of service distributions on system performance for the proposed voice and data model. For this purpose, the integrated voice and data system with a single buffer and a channel was considered and analysed using discrete event simulation. Mainly, four different distributions; exponential, normal, lognormal and weibull distributions were taken into account for service time, while packet arrivals followed a Poisson process. Simulation results revealed that service distribution has a great effect on system performance and weibull distribution fits well for the proposed model. Keywords Service time distributions, Performance analysis, Integrated voice and data services, Quality of service. 1. Introduction In network systems, the classification of service distribution is a significant requirement in service process for network administration, traffic engineering and customer experience. The number of wireless network users has been exponentially increased. This leads the increasing demand to provide seamless service, for not only voice but also data traffic. Therefore, along the service process, having different service characteristics affect the DOI: /TEM Corresponding author: Eser Gemikonakli, University of Kyrenia, Kyrenia, Mersin 10, Turkey eser.gemikonakli@kyrenia.edu.tr Received: 30 August Accepted: 12 October Published: 26 November Eser Gemikonakli; published by UIKTEN. This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivs 3.0 License. The article is published with Open Access at service distribution that need to be taken into account in designing a new network. Call admission control, buffer management and scheduling algorithms are mainly challenging tasks discussed in traffic management [1]. These all have a great impact on system performance. However, a deep understanding and analysis for the statistical distribution of service time is highly required for the proposed models. There are four service distribution mainly discussed in previous works. These are basically exponential, normal, lognormal and weibull distributions. The service time in exponential distribution is independent of the amount of service time completed so far. If the service time elapsed is a sum of several independent service times, the normal distribution can be considered and effectively used to build up service time model for the integrated services. Log-normal distribution is appropriate for modeling the service time if and only if the service time is a product of several independent service times. On the other hand, Weibull distribution for the service time can be considered in designing the heavy tailed systems. The huge amount of demand to use integrated services leads bursty traffic which may not be possible to explain their service processes with a simple Markovian model. According to the authors in [2], arrival process can be modeled as a homogenous Poisson process, with a fixed rate, while the distribution of the packet service time cannot be considered as exponential. For real case systems, it may not be possible to derive exact solution for the proposed service distribution. This is due to the mathematical intractability of the complex system. While the theoretical point of view in designing the mathematical model shed a light to system developers, it is also required to obtain the realistic performance analysis according to the vendor s perspective. Therefore, understanding data traffic statistics by analyzing real network traffic is important to network operators and other interested parties. Queueing and service policy for the integrated voice and data services are extensively used in 798 TEM Journal Volume 7 / Number 4 / 2018.

2 different ways in the literature. In most of them, guard channels are used to reserve different packet types [3-6]. According to the author's knowledge, this is the first study in the literature evaluating the effect of different service distributions on performance analysis for the integrated voice and data model, where a single buffer is considered instead of using reserved channels for different services to obtain the required quality of service (QoS). Therefore, this paper addresses the following two research questions. 1. What impact do the traffic service distributions have on the performance of the proposed voice and data model? 2. Which service distribution fits well with the proposed model in order to provide the desired QoSrequirements? To answer the questions raised above, discrete event simulation (DES) was developed and sensitivity analysis, for different service distributions were carried out to find the optimum performance results. The remaining parts of the paper were organised as follows: First, section 2 reviews the literature of most relevance to the research questions. Next, section 3 explains the proposed model in detail. Section 4 provides the simulation result. Section 5 summarises the study and section 6 discusses future research. 2. Theoretical Background The most general selection for service time distribution, in mathematical model design, is based on the exponential assumption [7], [8]. Generally, the arrival process is assumed to be Poisson and service time is considered to be exponentially distributed [9], [10]. However network traffic today is self-similar and its processes can exhibit long-range dependency [11]. This results bursty traffic over a wide range of time scales. In some of the studies discussed below, the data obtained from the experimental analyses were compared with the data generated by the statistical distributions. This observation and analysis provided the recognition of the service time distribution under heavy traffic load for the proposed network models. According to the author in [12], exponential distribution for call holding time did not comply with the experimental data. Their findings showed that the log-normal distribution fitted well with the empirical data. In [13], lognormal distribution was considered for the service times in call centre. Experimental data taken form mobile operator was used to determine the optimum service distribution in [14]. According to the authors, the data obtained by the lognormal-3 function represented the experimental data, where it was the variation of lognormal function. Simulation model was developed to analyse the performance metrics in [15]. In this study, service times for the integrated services were assumed independent and identically distributed with general distribution. The impact of general distribution on the steady state probability distribution was evaluated in [16]. They showed that higher-order properties of the service time distribution affected the steady-state probability distribution in M/G/1. According to the experimental analysis done for VoIP systems, the speaking and silence periods performed a Weibull distribution unlike the previous studies [17], [18], where both periods followed an exponential distribution. 3. The Proposed Model The proposed model, in Figure 1., offers a single buffer with a single channel instead of using reserved channels for different service types. The buffer is used for both voice and data packets and this provides optimum buffer threshold for integrated voice and data packets. The packets arrive to the system with mean arrival rates and for voice and data packets respectively. All the packets joined to the system are buffered in case the channel is not available. Buffer capacity is shown as for voice packets, while it is for data packets. is equal to the maximum number of data packets, if there are some voice packets in the system, where. Therefore, is the total number of voice packets in the system, where. Figure 1. The single buffer and a channel model The state transition diagram for the integrated voice and data service model is given in Figure 2. The lateral transition occurs from the state (i,j) to the state (i+1,j), when the voice packet arrives to the system. There is a backward transition from the particular state, from (i,j) to (i-1,j), when the voice packet is served. On the other hand, a data packet arrives to the system with an upward transition from (i,j) to (i,j+1), and if there is no voice packet in the system, service is given to the data packet at front of the queue and there is a downward transition from (i,j) to (i,j-1). TEM Journal Volume 7 / Number 4 /

3 4. Simulation Results In this section, simulation results are presented. The results show the behavior of the distributions and their impact on the system performance for the proposed model. The parameters used in the simulations are given in Table 1. Table 1. Summary of parameters used in simulation Figure 2. The single buffer and a channel model Let ( ) be the steady-state probability that simultaneously there are i voice packets and j data packets including the ones being served in the system, the corresponding balance equations for the model proposed are given as below. ( ) ( ) ( ) 0.24 packet/second Parameter Value λ v 0.22 packet/second λ d µ v [0.3-1] packet/second µ d [0.3-1] packet/second L v Ld S 1 Figure 4. shows the effect of on the for different distributions for service time. All distributions have a similar trend, and there is a decrease in as increases. Lognormal distribution has a reasonable effect on, which reduced the up to 0.081, when. The results for exponential distribution are greater than other distribution for various values. On the other hand, the normal and weibull distributions have almost same results for. For all distributions, increasing values further than 1 do not affect the system performance anymore. { In Figure 3., service policy of the integrated voice and data services is demonstrated. In case, there are some packets waiting in the queue, the voice and data packets receive service in a preemptive fashion according to the First Come and First Serve (FCFS) fashion. Therefore, voice packets have a priority over data packets. Data packets can take service if and only if there is no voice packet waiting in the queue. Figure 4. Figure 3.Service flow for the integrated voice and data packets is shown in Figure 5. for different distributions as increases. It is clear that, increasing reduces the voice traffic load, where values decrease. does not change for various values for lognormal distribution, while there is a sharp decrease in values for weibull distribution, when. 800 TEM Journal Volume 7 / Number 4 / 2018.

4 Figure 5. Figure 6. demonstrates for different distributions, but this time as increases. For all distributions, there is an increase in as increases. However, values for lognormal distributions are significantly greater than other distributions for various values. 5. Conclusion Figure 7. In this paper, performance metrics of the proposed integrated system were analysed based on different distributions of service time. According to the result obtained, lognormal distribution has a significant impact on for voice traffic service time. However, its effect on is not in reasonable range since the number of data packets blocked are over the expected value. A deep analysis was carried out and it was found that weibull distribution fits well for the proposed integrated voice and data model. Figure Further Work A more detailed literature survey will be carried out and the effect of service distribution, considering mobility issues, on system performance will be discussed. Figure 7. shows for different distributions as increases. decreases when increases for exponential, normal and weibull distributions, even for Lognormal distribution which provides a small amount of change for it. For weibull distribution, values are in the acceptable range ( ) for all values greater than 0.7. Normal and Exponential distributions satisfy this condition when. References [1]. Gemikonakli, E., Ever, E., Mapp, G., & Gemikonakli, O. (2017). Admission control and buffer management of wireless communication systems with mobile stations and integrated voice and data services. Telecommunication Systems, 65(4), [2]. Gualandi, S., & Toscani, G. (2018). Call center service times are lognormal: A Fokker Planck description. Mathematical Models and Methods in Applied Sciences, [3]. Haung, Y. R., Lin, Y. B., & Ho, J. M. (2000). Performance analysis for voice/data integration on a finite-buffer mobile system. IEEE Transactions on Vehicular Technology, 49(2), [4]. Ferng, H. W., & Tsai, Y. C. (2005). Using priority, buffering, threshold control, and reservation techniques to improve channel-allocation schemes for the GPRS system. IEEE Transactions on Vehicular technology, 54(1), TEM Journal Volume 7 / Number 4 /

5 [5]. Wang, J., Zeng, Q. A., & Agrawal, D. P. (2003). Performance analysis of a preemptive and priority reservation handoff scheme for integrated service-based wireless mobile networks. IEEE Transactions on Mobile Computing, 2(1), [6]. Mohamed, N. O. (2016, December). Performance analysis of combined guard channel and Mobile assisted handoff call admission. In Electronic Devices, Systems and Applications (ICEDSA), th International Conference on (pp. 1-4). IEEE. [7]. Silva, A. P., & Mateus, G. R. (2003). Performance analysis for data service in third generation mobile telecommunication networks. Journal of the Brazilian Computer Society, 8(3), [8]. Nan, E., Chu, X., Guo, W., & Zhang, J. (2013, August). User data traffic analysis for 3G cellular networks. In th International Conference on Communications and Networking in China (CHINACOM) (pp ). IEEE. [9]. Jain, M., Sharma, G. C., & Mittal, R. (2013). Performance analysis of prioritized call admission control schemes for integrated traffic in wireless network. International Journal of Electronics Communication and Computer Engineering, 4(1), [10]. Kokila, S., Shankar, R., & Dananjayan, P. (2012, March). Performance analysis of dual threshold call admission control in 3G/WLAN coupled network. In Advances in Engineering, Science and Management (ICAESM), 2012 International Conference on (pp ). IEEE. [11]. Tian, Y., Han, D., Liu, L., & Fu, Y. (2017, October). A self-similar traffic generation model based on time. In Microwave, Antenna, Propagation, and EMC Technologies (MAPE), th IEEE International Symposium on (pp ). IEEE. [12]. Gustafson, F., & Lindahl, M. (2009). Evaluation of statistical distributions for VoIP traffic modelling., 48. [13]. Liang, C. C., & Luh, H. (2015). Solving twodimensional Markov chain model for call centers. Industrial Management & Data Systems, 115(5), [14]. Pattavina, A., & Parini, A. (2005, August). Modelling voice call interarrival and holding time distributions in mobile networks. In Proceedings of the 19th International Teletraffic Congress (ITC'05). [15]. Ahmed, N., & Rikli, N. E. (2018). A QoS based Algorithm for the vertical handover between WLAN IEEE e and WiMAX IEEE e. Int. J. Com. Dig. Sys, 7(1). [16]. Brandwajn, A., & Begin, T. (2009, January). A note on the effects of service time distribution in the M/G/1 queue. In SPEC benchmark workshop (pp ). Springer, Berlin, Heidelberg. [17]. Papp, G., & GauthierDickey, C. (2008, June). Characterizing multiparty voice communication for multiplayer games. In ACM SIGMETRICS Performance Evaluation Review (Vol. 36, No. 1, pp ). ACM. [18]. Xi, B., Chen, H., Cleveland, W. S., & Telkamp, T. (2010). Statistical analysis and modeling of Internet VoIP traffic for network engineering. Electronic Journal of Statistics, 4, TEM Journal Volume 7 / Number 4 / 2018.