Traffic Management A Multi-Agent System Approach

This paper proposes an approach to the management of traffic in telecommunication networks. It was motivated by the growing need for intelligent network management in current telecommunication networks such as PSTN, ISDN or transmission networks. The goal is to constantly act upon the network in order to control the traffic flow, trying to maximize the number of calls that reach their destination or simply to reduce the congestion level of the network. In this approach, the traffic control task is based on a Multi-Agent System with learning capability. This control system depends on other peripheral systems, which supply processed traffic information (mainly performance alarms), interact with the user and forward traffic controls to the corresponding Network Elements. Each agent is dedicated to the control of a single Network Element. The whole structure of the Multi-Agent System will control the communication network. In order to provide adaptation and evolution of the control strategies each agent has on-line learning capability with an algorithm combining concepts from Cobweb and Classit. The subject of this article is closely related to one of the services of the Telecommunication Management Networks (TMN) known as Traffic Management. The TMN management services are specified in a series of recommendations produced by an international organisation called International Telecommunications Union (ITU) (previously CCITT). The main goal of ITU is the standardisation of a set of services in the area of telecommunication networks (TNs). The Traffic Management (TM) service is devoted to the traffic management of circuit switched networks such as Public Switched Telephone Network (PSTN), Integrated Services Digital Network (ISDN) and transmission networks such as Synchronous Digital Hierarchy (SDH). The main purpose of this service is to maximise the number of successful calls across the network (NW) (and/or to maximise revenues, e.g. increase the number of expensive international/national calls sacrificing local calls). This goal is achieved through the maximisation of the usage of allavailable equipment and facilities under any condition. The TM service is also seen as a NW performance monitor, allowing whenever necessary, the possibility to take appropriate actions in order to control the traffic flow, in such a way that the overall NW performance is enhanced [E411-IT92]. In a first attempt and to assure an easier service specification, the TM service is oriented towards the Network Element (NE), here assumed to be the digital exchange. The TMN Operations System (OS) collects traffic information from the NEs, analyses it and sends them commands to change their operation or to change the NW configuration. The NEs may report the traffic information on a periodic basis or triggered by some threshold value. The TMN OS may change those threshold values and/or the periodicity of the traffic information reports. The data delivered by the NEs may be processed within the TMN, via Mediation Devices or by the OS. The OS must collect all the relevant TM indicators in order to get an overall NW status information. 1 CENTRIA Departamento de Informática, Faculdade de Ciências e Tecnologia, Universidade Nova de Lisboa, Portugal. Figure 1 shows the main flow of data and control between the monitored NEs and the corresponding OS. The unit we are focusing in this work is the rectangular box labelled “Intelligent Traffic Control”. The OS may influence the NW by issuing traffic volume controls and traffic routing controls. Note, however, that the TM is done at NW management level and not at NW level. This is an important issue, since the NW may be heterogeneous, i.e. with NEs from different vendors. There are other traffic control techniques, which try to deal with some of the problems described in this paper, but which are solved at NW level. For instance some NEs support automatic congestion controls, dynamic call routing [OB96] or allow different adaptive routing algorithms [CD97]. The dynamic TM described in this paper is specially useful in adverse traffic conditions, e.g. mass call situations caused by natural disasters, public events, etc., since these kind of problems need a quick response time from the NW management operators. This system may be proactive or reactive in the TM, however the main focus will be put in the reactive TM. The main motivation for using a Multi-Agent System (MAS) is the great complexity of the resources being managed, i.e. the NW, its distributed nature and the amount of information to be processed and correlated. The control task aims to achieve a balance between two contradictory goals: the maximisation of the overall NW performance, e.g. increase the number of successfully handled calls and the minimisation of the congestion level which reduces the performance degradation, e.g. reduce the traffic volume. In NWs with a large number of NEs, a solution based on a single global agent becomes too complex to be handled. Since the NW traffic may only be controlled at NW nodes, it seems natural to provide each node with a Control Agent (CA). Furthermore, this association enables the CA to adapt itself to the specificity of the managed NE, for instance the NE type (transit, local, international, etc.) or the NE vendor (which may impact on the NW management controls available). Another advantage of this architecture is the use of dedicated Data Communication Network (DCN) links (see section 5) which protects the MAS from congestion problems in the NW Intelligent Traffic Control Traffic indicators Control actions