مدیریت کیفیت در شبکه های GPRS با استدلال فازی مبتنی بر مورد

Mobile networks pose challenges to quality management because of the limited capacity of the air interface and the mobility of the users. GPRS is the prevailing system for mobile connectivity at the moment. This paper approaches quality management in GPRS networks with a two-phase system, where a detector block first culls quality disturbances and a fuzzy case-based reasoning engine then proposes a solution to the problem. The main advantage of the concept is model maintenance: the experienced network operator can take part in the decision-making and his or her knowledge thus accumulates in the case base. We also present simulated examples of GPRS network data classified with the detector and inserted into the case base.

The rapid growth of telecommunications has resulted in an ongoing trend towards more complicated network systems. The growth is not always controlled; expanding and connecting subsystems of latest technology as well as legacy networks brings along management problems. Problems in the network give rise to a need of fault management. A fault is a disorder occurring in the hardware or software of the managed network [1]. Fault management in turn is defined as a five-step process [2]: fault detection, fault location, service restoration, identification of the problem’s root cause and problem resolution. The above definition of fault management involves fault detection and location. Fault isolation and localization are further terms used in the context of fault management with more or less different shades of meaning [3]. Reported approaches to these tasks include neural networks [4], belief networks [5] and graph theory [6], just to name some of the recent ones. An extensive review is provided in [3]. In addition to faults, we address performance management in this paper. Performance management [2] entails problems that are not actual faults in hardware or software but may appear as faults to the user, such as excessive amount of users in a certain network segment. As these problems often pertain to the conception of quality, we use the term quality disturbance in this paper to denote all quality degradations originating from different causes. Quality of Service (QoS) relates to performance management in the sense that QoS attempts to rescue the quality of the connection by dividing it into different quality classes, such as gold, silver and bronze. One of the main problems of QoS is measuring quality, particularly the quality perceived by the end user [7]. A variety of QoS parameters [8] can be measured from the network to describe the quality. We use statistical parameters to define acceptable QoS for the quality management system. General Packet Radio Service (GPRS) is a service enabling mobile access to the Internet. In contrast to circuit-switched mobile communications, GPRS uses packet-switching, which results in effective use of radio bandwidth. This has made GPRS a system affordable to a mass market. Whether or not third generation networks will truncate the life cycle of GPRS, remains to be seen [9]. Inherent problems with mobile network QoS include link quality, mobility of the user and limitations of the portable device [10]. Combining these with the complexity of the wired network poses challenges to the quality management of GPRS networks. While some problems in the wired world can be solved simply by adding bandwidth at moderate expense, the air interface has a physical limit. Steinder and Sethi [3] declare fault localization in mobile networks an open research problem. Indeed, very few publications cover the particular problem of GPRS networks and fault management. Yoneki and Bacon [11] introduce semantics for event correlation in complex mobile networks. In this context, event correlation is a term closely related to fault localization. Kant [12] discusses self-healing properties of wireless networks and proposes a restoration mechanism combined with QoS policies for GPRS network. Lewis has published a book on network management and case-based reasoning in 1995 [13] but at that time there were no mobile packet data networks. Thus, our fuzzy case-based reasoning approach to GPRS network quality management can be termed novel. The results presented here are a part of a larger research project [14]. The focus of the project was on analysis and management of networked systems by means of intelligent data analysis. Complex and changing networks call for sophisticated methods combined with the traditional tools of the very application. This paper concentrates on one such application, GPRS network, and combines an adaptive performance management system with the existing GPRS network management. We introduce a two-phase quality management concept that consists of a detector and a fuzzy case-based reasoning (FCBR) block (Fig. 1). The detector observes measurements collected from the network and separates disturbances from normal operation. As soon as the detector detects a disturbance, it creates a trouble ticket[15] and passes it to the FCBR block. The trouble ticket contains all measurements at the moment as well as a number identifying the disturbance class.When the fuzzy CBR engine gets a trouble ticket, it checks its case base to see whether an existing case matches the current one. This is where fuzziness comes in: the operator gets proposals with different membership grades that indicate the degree of match. The final decision of which case to choose is up to the operator. Most importantly, the operator also has the possibility to give feedback to the FCBR engine. Hence the case base evolves to fit the changing needs of the whole system. While the system can adapt to conform to many kinds of fault management problems, here we refine and apply it with a GPRS network in mind. The main advantage of the concept is model maintenance: the knowledge of an experienced user inherently helps in adapting to new and changing situations. Additionally, keeping the detector and FCBR blocks separate lets one train the detector off-line and adapt the FCBR during online use. The operator needs no knowledge on the underlying algorithms or programming, he or she just provides the expertise. The system has two major modes, off-line training and online operation. Section 2 concentrates on the off-line mode, that is, training a supervised algorithm. After the detector has been trained, Section 3 describes how detector and FCBR blocks work together online. We also present some test cases for the concept with a fluid flow model for the GPRS network in Section 4.

We have presented a new approach to GPRS network quality management. The proposed system contains two parts, a detector and a fuzzy case-based reasoning (FCBR) engine. Using two independent blocks is one of the strengths of the system. Thanks to the detector, the operator need not extract disturbances from the vast information flow by himself. Furthermore, the method of the detector may be chosen or even changed independently of the FCBR part. Model maintenance is a common problem in changing environments. The solution to this problem is inherent in our system. The operator has the possibility to influence each case, thus his or her expertise gradually accumulates into the case base and the system evolves to cover the current state of the network. In the above, we considered the distinction of detector and FCBR an advantage, but it may also have negative effects. The detector is static, that is, it is not updated as the case base is. Naturally, it can be retrained occasionally, but retraining in turn may cause incoherence with the case base. One way to fix the case base after retraining the detector is to update all cases individually to see if they have fallen into a different disturbance class (see Fig. 4). The case-based reasoning block also has potential for further development. Case adaptation techniques [13] could help in creating new cases. For example, if a disturbance has occurred for the silver traffic class, the CBR engine could find a similar case for the bronze class and adapt it to the present case. Or a parametric solution might suggest to increase the core network bandwidth by kN when the amount of users is N.