Complex Application Architecture Dynamic Reconfiguration Based on Multi-criteria Decision Making

 Internet & Technology

of 19
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
International Journal of Software Engineering & Applications (IJSEA)
  International Journal of Software Engineering & Applications (IJSEA), Vol.1, No.4, October 2010DOI : 10.5121/ijsea.2010.1402 19 Complex Application Architecture DynamicReconfigurationBased on Multi-criteria Decision Making Vincent Talbot (1) & Ilham Benyahia (2) Université du Québec en Outaouais, 101 St-Jean-Bosco, Box 1250, Hull Station,Gatineau, Québec, Canada J8X 3X7  ABSTRACT   Intelligent Transportation Systems (ITS) are increasingly important since they aim to bring solutions tocrucial problems related to transportation networks such as congestion and various road incidents. Management of ITS, as other complex and distributed applications, has to cope with unforeseeable eventsand incomplete data while guaranteeing a quality of service (QoS) defined by multiple criteria reflectingreal-life needs. To enable applications to adapt to changing environments, we define a methodology of dynamic architecture reconfiguration based on multi-criteria decision making (MCDM) using evolutionarycomputing (EC) to find the best combination of architecture components. We use the Pareto Evolutionary Algorithm Adapting the Penalty   (PEAP), a category of EC, selected in this paper to deal with time-consuming online processing required by basic EC such as genetic algorithms. Our simulation resultsrelating to road safety highlight the benefits of MCDM prior to such reconfiguration. We also address the problem of destabilization which can result from repeated reconfigurations in response to ongoingenvironment changes.  KEYWORDS Complex applications, architecture performance optimization, architecture reconfiguration, multi-criteriadecision making, Pareto Evolutionary Algorithm Adapting the Penalty   (PEAP), road safety application. 1.   INTRODUCTION Emerging technologies in computing and telecommunications have brought valuable newdimensions to the development of complex applications in many domains including intelligenttransportation systems. Traffic and routes are continuously monitored, and reports on their currentstates are transmitted to a central station where all the information is visualized and analyzed.Then, control data are transmitted to different locations within a specified deadline so that trafficsystems can be adjusted according to new road conditions to avoid road congestion and to help  International Journal of Software Engineering & Applications (IJSEA), Vol.1, No.4, October 201020 drivers restrict their speed to avoid accidents. The software supporting the control processingrequires an architecture that can be reconfigured in response to changing road conditions. Toensure that the processing components of a traffic management system are readily adaptive, ourapproach is built around dynamic reconfiguration of the software architecture. To support suchreconfiguration, we set out a new approach to multi-criteria decision making (MCDM) based onthe Pareto Evolutionary Algorithm Adapting the Penalty ( PEAP) to improve upon the traditionalslow genetic algorithm process. To reach appropriate final decisions related to reconfiguration wepropose the integration of supervised learning or interactions with a decision maker to find thesolution that best satisfies the specific objectives of the application being considered with PEAP.We also address the important problem of destabilization, which is a serious threat when frequentreconfigurations occur.This paper is organized as follows. Section 2 provides background on related work involving theuse of reconfigurable software architecture. Section 3 gives an overview of MCDM incorporatingmetaheuristics with a focus on transportation applications. Our approach to dynamicreconfiguration based on evolutionary techniques is detailed in Section 4. In Section 5, we presenttwo scenarios specific to road safety illustrating the benefits of dynamic reconfiguration based onmulti-criteria decision making.   Section 6 presents a conclusion and future directions. 2. Related Work Much recent research on distributed systems that interact with their environments has focused ondynamic reconfiguration and adaptive resource management [1], [2], [3], as means of optimizingand guaranteeing the required quality of service (QoS). However, premature triggering of reconfiguration may result in system instability and performance degradation because of theuncertainty created by frequent changes in the operating environment. To maintain good   operation equilibrium, the tendencies of these systems must therefore be evaluated before thereconfiguration process is launched. 2.1 Dynamic Reconfigurable Frameworks Dynamically reconfigurable systems typically incorporate component-based frameworks capableof modeling, managing and reorganizing their architecture with little or no human intervention.OpenRec [4] and Fractal [5], [6] are two such frameworks capable of introspection andextensibility. Both have recently been extended with formal specification metalanguages (Alloyand Focal, respectively), used to model systems and to prove dynamically that systemsconfigurations are semantically correct and satisfy functional constraints.The drawback of these reconfigurable systems is that they lack processing capability to deal withinstability associated with frequent reconfigurations. Sophisticated tools such as UML-RT [7]designed for specification validation of real-time distributed systems are of limited value fordynamic validation of adaptive systems that interact with disturbed environments. Simulationframeworks are more appropriate for that purpose since various scenarios can be tested forperformance validation. In this paper, we use a real-time, process-driven simulation environmentcalled J-Sim [8] to evaluate the performance of each architecture configuration.  International Journal of Software Engineering & Applications (IJSEA), Vol.1, No.4, October 201021 2.2 Management of Architecture Solutions Based on the Reconfiguration In order to cope with changes in its environment, a self-adapting system must locate, discover orconstruct alternative configurations and select the most appropriate one for the current   environment context according to QoS criteria such as deadline satisfaction and operating rate.The approach proposed in [9] is based on generating reactive plans (configurations) from goalsexpressed in temporal logic. Its three-layered conceptual model consists of a goal managementlayer, a change management layer and a component layer. In [10], alternative solutions for agiven application domain are captured in a domain repository. Similar approaches are found inthe field of cognitive radio[11]where hierarchical management is used to discover and construct dynamically new hardware and software architectures for cognitive radio systems. In [12] ageneric cognitive framework is presented for autonomous decision making. Multiple possiblyconflicting, and operational objectives are analyzed in a time-varying environment.Our focus here is on the impact of dynamic reconfiguration on ITS applications, specifically thoserelating to the comfort, safety and security of motorists. As shown in [13], some traffic modelsmay lead to congestion that can cause crashes. Clearly, one objective of reconfigurations must beto balance traffic loads. Another essential objective in traffic monitoring reconfigurations must beto enhance dynamic emergency vehicle dispatching systems. 3. Multi-criteria Decision Making (MCDM) for DynamicReconfiguration Just as the configuration of architectures intended for complex applications including real-timesystems must be validated, so the reconfiguration of these architectures must take into account theQoS parameters to be validated online. When the QoS is defined by a set of criteria and there arenumerous alternatives for architecture reconfiguration, MCDM is essential. 3.1 Evolution of MCDM Pioneer research in multi-criteria decision making (MCDM) dates back to the 1950s. MCDMthen evolved in two main methodological directions, the first based on an outranking criteriaapproach to deal with heterogenous criteria and their associated scales [14], [15] and the secondon multi-attribute utility theory [16].Today, new approaches to MCDM take advantage of advances in information technologiesapplications to solve both theoretical and applied decision problems such as those encountered inITS. One frequently used method is to reduce multiple criteria to a single criterion byaggregation. However, there is little research on combining multiple objectives in a more realisticway, especially when there is an absence of dominance or potential conflict between two or morecriteria. Recent work on MCDM in the field of transportation has focused on solution researchwithin large spaces of feasible solutions and involves the application of metaheuristics andlearning techniques such as supervised learning, a category of evolutionary algorithms (EA) suchas genetic programming [17].  International Journal of Software Engineering & Applications (IJSEA), Vol.1, No.4, October 201022 3.2 Use of Metaheuristics in MCDM for ITS Increasingly, multiple-objective metaheuristics is being applied to the solution of complex ITSproblems such as vehicle assignment, routing and scheduling or crew assignment and scheduling. In this context, there is a pressingneed to overcome major challenges in MCDM techniques foruse online to manage highly frequent environment changes.Many EAs deal with optimization problems complicated by performance requirements. One suchexample is the Strength Pareto Evolutionary Algorithm (SPEA) [18] which is based on therelative strength of individual solutions within a solution space. However, when used for complexapplications with QoS requirements, MCDM optimization must be constrained by performancerules as well as semantic rules (e.g., correctness of the composed architecture).Little research has addressed MCDM problems in dynamic and constrained applications. Themain issue is that problem solving methods such as evolutionary techniques require numerousiterations to find an optimal combination yet there are temporal constraints for making online   decisions that must be met. In [19], MCDM is applied in the dynamic reconfiguration of opticalnetworks characterized by very high speed transmissions. An approach based on ParetoEvolutionary Algorithm Adapting the Penalty   (PEAP) which is characterized by individualpenalties is used in this context. The main purpose of multi-criteria optimization is to find thePareto border defined by a set of potential solutions. In our work, we have therefore employed ahybrid technique that combines Pareto-optimal sets with penalty functions. 4. An Architecture for Dynamic Reconfiguration Incorporating MCDM We outline here a methodology of architecture reconfiguration based on evolutionary techniquesand incorporating MCDM. While the approach presented in this paper has been developed in thecontext of ITS, it is broadly applicable to complex applications that must meet QoS requirementswhile operating in a disturbed environment.Our network management framework [20] centers around a set of software agents called ComplexAgents (CAs) that are equipped with functions to monitor and control their environment. They are   autonomous software entities which are capable of reconfiguring their internal processingarchitecture in order to maintain an optimum level of control and monitoring despite unforeseenchanges in their environment. The architecture of a CA is based on a library of softwarecomponents classified according to the processing cycle of environment events. Differentcomponents may serve the same end purpose, but with different performances. For instance,scheduling of real-time events can be carried out by a number of different algorithms such as firstdeadline, earliest deadline, time laxity, etc. [21].The CAs rely on a sophisticated architecture reconfiguration management (Fig. 1). Parametersrepresenting the performance criteria of each CA component will always include processingaccuracy and the CPU execution time.
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks