Noriteru Shinagawa

A proposal for a mobile communication traffic forecasting method using time-series analysis for multi-variate data

In the present mobile network, the management of the radio resource to connect users and edge networks is an important issue. If there is a rapid rise in traffic coming into the network, regulation methods such as admission control or admission avoidance are the only way of dealing with this. Since it is beyond the mobile network systems capability to predict fluctuations in traffic for every area, there are great difficulties in realizing other control methods. In order to solve these problems, this paper proposes a method of forecasting traffic systematically based on the users properties and information about the environment.

Diffusion Approximations for the GI/G/c/K queue

Although diffusion approximations are widely used for analyzing non-Markovian queueing systems, so far there are few results on the relationship between the approximation accuracy and a combination choice of two boundary conditions and two discretization for finite-capacity multi-server queuing systems. The finite-capacity system requires two boundary conditions at origin (x=0) and capacity limit (x=K). Taking account of this requirement, four types of diffusion approximations have been presented. We present accuracy comparisons among four types of diffusion approximations which differ from one another in the boundary conditions and the discretizations. We observe that the coefficient of variation of the inter-arrival time distribution is a key factor for choosing the appropriate combination of the boundary conditions and the discretizations.

Performance Evaluation of Network Systems Accounting for User Behaviors

Considering user behaviors in the performance evaluation of mobile networks is crucial as traffic generation in such networks is highly dependent on mobility and communication activities of mobile users. This paper presents a simulator that can faithfully simulate user behaviors together with the target network system. The following three components have been implemented in a network simulator in order to model realistic user behaviors: (i) a function of loading geographical data for the target area, (ii) a pedestrian mobility model in which users walk along streets based on the given geographical data, and (iii) a communication behavioral model, which accounts for correlations between user profiles and their communication activities. Simulation of a mobile network system is performed and shown that the simulator replicates realistic behaviors of individual users, and also that the user behaviors do influence the predicted overall system performance. This indicates the importance of accounting for user behaviors in mobile network system evaluation.

THE ROBUSTNESS OF BLOCKING PROBABILITY IN A LOSS SYSTEM WITH REPEATED CUSTOMERS

In this paper, we analyze and synthesize a multi-server loss system with repeated customers, arising out of NTT DoCoMo-developed telecommunication networks. We first provide the numerical solution for a Markovian model with exponential retrial intervals. Applying Little’s formula, we derive the main system performance measures (blocking probability and mean waiting time) for general non-Markovian models. We compare the numerical and simulated results for the Markovian model, in order to check the accuracy of the simulations. Via performing extensive simulations for non-Markovian (non-exponential retrial intervals) models, we find \(\it{robustness}\) in the blocking probability and the mean waiting time, that is, the performance measures are shown to be insensitive to the retrial intervals distribution except for the mean.