Abstract
Сontents
- Introduction
- 1. Theme urgency
- 2. Goal and tasks of the research
- 3. Application of the ARFIMA (p, d, q) for the prediction multiservice traffic
- Conclusion
- References
Introduction
Modern telecommunication networks are characterized by rapid growth in traffic. According to the analytical group TASS-Telecom for the last 5 years, the global market of information technology are growing annually by 10% [1]. Cisco Visual Networking Index shows that, IP-traffic will increase by 29% from 2011 to 2016 [2]. Recent years there is a tendency towards convergence of telecommunications services, which creates additional difficulties in its transmission and processing. Multiservice traffic requires prioritization, and each class makes certain quality of service requirements. Due to this it is necessary to provide flexible system of network control by dynamically reallocate resources of communication channel (bandwidth, transmission speed, etc.). It is necessary to take into account some of the features of modern multiservice traffic. The traditional load balancing algorithms can be significantly enhanced through the application of predictive control. In this work will be affected long-term forecasts that in the future will allow provide the strategic planning of telecommunication networks in terms of the efficient use of its resources and to avoid congestion, leading to the loss of information.
1. Theme urgency
For a long time it was thought that the traffic data network can be described by the classical Poisson distribution. However, in recent years, studies have found that high-speed network traffic has spikes in demand which lead to network congestion. That is, the classical methods of calculating networks impractical for these conditions. We ourselves can trace the changes, for example, the load on the mobile network at different times of day, on different days of the week. All this indicates that the traffic is self-similar in nature.
"Self-Similarity" is a property of the process to maintain their behavior and external signs of the consideration of different scale [3]. Self-similar process looks less smooth, more uniform (ie, has a higher variance) than a purely random process.
2. Goal and tasks of the research
Improving the quality of the provision of telecommunications services with an efficient use of network resources by using the predictive model ARFIMA is the goal of research.
Main tasks of the research:
- Analysis of the features of modern telecommunication networks and the load.
- Statement of traffic requirements to be forecast.
- Select the method by which the forecast will be realized.
- Determine the order parameter selection predictive model.
- A study to forecast and assessment of the effectiveness of the method chosen.
Research object: processes in telecommunications networks.
Research subject: telecommunication network traffic.
As part of the master's work is to get the actual scientific results: It is planned to develop predictive ARFIMA-model capable of adequately describe the processes in telecommunication networks with the property long-term memory.
3. Application of the ARFIMA (p, d, q) for the prediction multiservice traffic
The use of prediction methods allows to know in advance what kind of bandwidth will need to allow for maintenance of incoming traffic. There are several generally accepted methods for time series prediction. Methods for Box-Jenkins (ARIMA), in contrast to other methods of forecasting, do not provide any clear model for the prediction of the time series. Set only a general class of models that describe the time series and allow a certain way to express the current value of the variable in terms of its previous value. The algorithm then, adjusting the internal parameters, chooses the most suitable prediction model. ARFIMA approach allows to predict long-term memory processes, i.e. can be used to predict the traffic, having sufficient correlation between the time remote samples. For the purpose of this method is estimated stationary process, if necessary, a number of preorazovyvaetsya to a stationary form, and then set the parameters of AR component moving average (the behavior of the normalized ACF) and fractional intergrated component (using Whittle) [3, 4].
Conclusion
The properties of modern multiservice traffic that allow for long-term forecasting. Of the many methods of forecasting methodology chosen Box-Jenkins as the most modern, relevant approach to the study of time series.
During writing the abstract Masters qualification work is not completed. Final completion: December 2013. Full text of the materials on work can be obtained from the author or his supervisor after that date.
References
- Two traces contain two month's worth of all HTTP requests to the NASA Kennedy Space Center WWW server in Florida [электронный ресурс]. – Режим доступа: http://ita.ee.lbl.gov/html/contrib/NASA-HTTP.html.
- Cisco Visual Networking Index: Forecast and Metodology, 2011-2016 [электронний ресурс]. – Режим доступа: http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white_paper_c11-481360_ns827_Networking_Solutions_White_Paper.html.
- Костромицкий А.И. Подходы к моделированию самоподобного трафика / А.И. Костромицкий, В.С. Волотка // Восточно-Европейский журнал передовых технологий. – 2010.
- Перцовский О.Е. Моделирование валютных рынков на основе процессов с длинной памятью: Препринт WP2/2004/03 – М.: ГУ ВШЭ, 2003. – с. 15-17.