Understanding traffic characteristics plays a vital role in network protocol design, which aims to minimize network resource consumption and maximize network stability. In this paper, we use the example of the mobile instantaneous messaging (MIM) service in cellular networks and try to understand its traffic characteristics. Specifically, in order to reach credible conclusions, our research uses practical measurement records from the MIM services of China Mobile at two different levels. First, a dataset of individual message level (IML) traffic is examined and reveals power-law distributed message length and lognormal distributed inter-arrival time. The heavy-tailness of which completely diverts from the geometric and exponential models recommended by 3GPP. Second, another dataset is used to examine the statistical patterns of aggregated traffic within a base station, and demonstrates the accuracy of $\alpha$-stable models for aggregated traffic. Furthermore, we verify that $\alpha$-stable models are suitable for characterizing traffic in both conventional fixed core networks and cellular access networks. Finally, by using the generalized central limit theorem, we build a theoretical relationship between the distributions of IML traffic and aggregated traffic.
(Color online) The spatial traffic density of mobile instantaneous messaging service in a random selected region with 23 active base stations. (a) 4:00 a.m.; (b) 10:00 a.m.; (c) 4:00 p.m.; (d) 10:00 p.m.
(Color online) (a)$\sim$(d) The accuracy after fitting the aggregated traffic in four randomly selected base stations by $\alpha$-stable models; (e) the PDF of fitting $\Psi(\omega)$ with respect to $ \ln (\omega) $ by using a linear function; (f) the PDF of fitted $\alpha$ values
Table 1 The modeling accuracy in terms of RMSE for Wechat
Name
PDF
Message length
Inter-arrival time
Aggregated traffic
Power-law
$ax^{-b}$
9.76E$-$5
9.25E$-$5
0.0357
Geometric
$(1-a)^xa$
607E$-$5
48.0E$-$5
0.0258
Exponential
$a{\rm e}^{-bx}$
56.0E$-$5
22.9E$-$5
0.0899
Weibull
$abx^{b-1}{\rm e}^{-ax^b}$
65.8E$-$5
8.08E$-$5
0.0470
Lognormal
$\frac{1}{\sqrt{2\pi}bx}{\rm e}^{-\frac{(\ln x -a)^2}{2b^2}} $