Article Preview
Top1. Introduction
One of the typical tasks solved by the TN monitoring and control systems is the network reconfiguration based on the observed operating conditions and options for their possible changes (Krinkin et al., 2021).
The reason for network reconfiguration may arise due to the failure of one or more network elements, an increase or decrease in the activity of network resource consumers resulting in changes of traffic volume as well as under the influence of other factors the combination and relationship of which cannot be clearly described. Timely reconfiguration allows avoid out-of-service situations for network users thus ensuring the required level of quality of services provided to them. The reconfiguration ensures balance and distribute of the load on network equipment and communication channels preventing the equipment from overload and maintains the ability to switch to standby capacities in case of emergency (provision of capacity reserve). To reconfigure networks, it is necessary to be able to forecast their states including network traffic based on data received from the TN devices.
In modern telecommunication networks the traffic volume commonly consists of incoming and outcoming Internet traffic of subscribers’ devices. Traffic depends on many factors, the one of the most significant is users’ behavior varying with the time of a year and of a day either working or weekend, the groups of subscribers and users’ tariffs, events that may affect the consumption of network services by subscribers, etc. The volume of traffic transmitted by Internet networks is growing with the increase of device structure complexity and users’ behavior. The number of smart devices such as set-top boxes (STB), smartphones and personal computers with installed intelligent client applications of TN operators like messengers, video content viewers, audio and video conferences, etc. has increased.
In TNs which include complex intelligent devices the end-users services assume interaction of those devices. The interaction can be organized according to various scenarios both given a priori and generated in dynamic mode. For example, when providing a VOD (Video on Demand) service to a network subscriber the STB communicates with several back-ends (CDN with video content, WEB server that provides a video catalog UI, authorization server, billing system, etc.). At the same time user behavior as regards the use of the service can be different (i.e., viewing the directory, watching one video in multicast mode, often started watching different videos, etc.) thus producing different load at back-ends.
For example, frequent switching of TV channels results in an increased load on the Fast Channel Change (FCC) server (Lin et al., 2009). Therefore, user commands’ analysis is an important component for network state forecasting. There are a number of models and methods intended to solve forecasting problem for TNs, for example, (Krinkin et al., 2021) proposes a solution using TN state models presented in the form of knowledge graphs. These models can be successfully used if the TN operator does not need to know which devices are involved in the provisioning of telecommunications services and receive/transmit data, and only the amount of the received and/or transmitted traffic is required. The models proposed in (Krinkin et al., 2021) are successfully used under conditions where traffic is either independent of user actions or the set of possible user actions is very limited.
In case when it is necessary to reconfigure the network to avoid sharp increase in network traffic in a certain location or a lack of performance of one of the back-ends caused by the behavior of groups of users, it is necessary to develop an extended model based on the model proposed in the (Krinkin et al., 2021). The model is to be extended by adding data about the network behavior, in particular, data about the behavior of each of the devices and the connections between them being set in the process of performing requests of users while providing them the telecommunication services. The model used in (Krinkin et al., 2021) is built using multilevel synthesis, the general theory of which has been significantly developed and by now is widely used in practice (Bardini et al., 2017; Courgeau, 2007; Osipov et al., 2019; Zhukova, 2019).