(IJCSIS) International Journal of Computer Science and Information Security,Vol. 9, No. 7, 2011
presence of congestion; and the effect of handover occursfrequently etc.The second important parameter for mobile cellular network planning is the channel holding time. It can be defined as thetime during which a new call occupies a channel in the givencell, and it is dependent on the mobility of the user. In the past,it has been widely assumed as the negative exponentialdistribution to describe the channel holding time .The probability of holding a call by a further time dt after holding the call up to time t isThe hypothesis of negative exponentially distributedchannel holding time is valid under certain circumstances .The channel holding time has been also been showed to fitlognormal distributions better than the exponential one .Also, several other works are also contradicted this simpleassumption. In [9,10] the probability distribution that better fits empirical data, by the Kolmogorov-Smirnov test, wasfound to be a sum of lognormal distributions.In some other works, it is shown that the channelholding time is also affected by user mobility. It ischaracterized by the cell residence time i.e. period of stay of acall in a cell. The cell residence time also follows definitedistribution pattern. The channel holding time distribution wasderived analytically [11, 12, 13] when the cell residence timehas Erlang or Hyper-Erlang distribution. A further empiricalstudy on GSM telephone traffic data reported in  whereanswered call holding time and inter-arrival times were foundto be best modeled by the lognormal-3 function, rather than bythe Poisson and negative exponential distribution.All the studies thus could not unanimously declare the bestoption between the classic Poisson model and the exponentialmodel for telephone traffic in cellular networks. In contrast,they suggested that call arrivals and holding time distributionmay be significantly time-correlated, due to congestion, user mobility and possible correlation between neighboring users.Study of all previous work lead us to further investigate theexact correlation of recent mobile traffic behavior with classicmodels and to check whether the traffic characterizationobtained would follow the previous behavior and models. Also,as a step ahead, if classical models are applicable as best fit,then the extent of percentage variation applicable for actualtraffic data.III
RAFFIC CHARACTERIZATION AND
NALYSIS OF TRAFFICDATA SETS
.In a Mobile network, traffic refers to the accumulatednumber of communication channels occupied by all users. For each user, the call arrivals can be divided into two categories:incoming calls and outgoing calls. Since every incoming callfor one user must be originated from an outgoing call of another user, we only need to consider outgoing calls fromeach user when we analyze the network traffic. Therefore, if not specified, we consider all outgoing calls as call arrival inmobile network for analysis purpose.All Outgoing calls are initiated randomly; if a call arrivesand the communication is successfully established, both thecaller and the receiver will be engaged for certain duration.The duration of the holding time is also a random variable.Thus, the traffic load depends on the rate of call arrivals andthe holding time for each call. Generally, Trafficcharacteristics of mobile network are typically measured interms of the average activity during the busiest hour or peak hour of a day.This paper presents a design approach to characterize themobility related traffic parameters in the presence of realtraffic conditions in urban area and rural area base on Cellcoverage. This includes the distribution of the arrival processes and the channel holding time.We analyzed sets of GSM telephone traffic data, collectedfor billing and traffic monitoring purpose which include callarrival time i.e. (Termination point of call) and the duration of calls at particular cell site. In addition, we also consider trafficother then voice calls like SMS service which may also affectthe network performance. Un-answered calls attempt could not be recorded and also no information was recorded to trace theuser mobility between the cells, neither was they felt necessary,as totality of the calls were recorded and attributed to theoriginating cell.All unsuccessful repeated call attempts, the impact of handovers and congestion were not taken into considerationfor present analysis. The different graphs have been plotted tofind the relation between the actual data and the classicalmodels.[A].
Analysis of peak traffic
We plot the graph of total traffic offered in erlangs at eachcell site. We had considered scale is discrete with one hour intervals to find the number of peaks occurs during the 24hours intervals. Next, we have calculated the average trafficload, peak hour load and the peakdness factor to find the trafficvariation and peakdness range for given number of channels. Inour calculation, peakdness factor has been defined asIdeally the value of peakdness factor lie within the range of 1 to 5 .Greater the range of peakdness factor means thatserver is over utilized and there may be chance of call drop.Total traffic characteristics depend upon actual traffic loadcarried by the server. This carried load consist of traffic other then voice service like SMS originated; which also affect theutilization of server performance. As a result it is important toevaluate the rate of the SMS service to predict the behavior of mobile users along with performance. Also, now a days,several companies offer bulk messages delivery in slack hour atvery cheap cost. As a result, number of users may use thisservice at redundant which may affect the quality of the voiceservice provided by the operators.