Uncertainty-oriented Runtime Service Decision-making Optimization Method

During the execution of the service process, service system and the externalenvironment facing a variety of uncertain events, so that the service cannot continuenormal execution, or the value cannot meet the user’s expectations. On Web services,reflected in the client program or server program fails, or cannot be normal interaction,resulting in one or more of the services cannot be successfully executed, thus affectingthe entire service process is not working properly. On business services, reflecting thechange in the demands of the customer or the service provider’s resources, cannot fullymeet customer’s demands in a service activity, so that the entire service process is notworking properly. To address this uncertainty event, to find the optimal decision-makingaction during the execution of the service is needed, so that both the loss ofuncertainties and the cost paid to selected strategy are minimized.For web services and business services, services uncertain events are classified anddecision-making scheme is analyzed under different uncertain events. Web service andbusiness service execution process are formalized. Actions for uncertainties of differentlevel are analyzed. Next, the profits and cost actions are calculated. Respectively, usingMarkov decision process (MDP) and the greedy algorithm to select the optimaldecision-making action.(1) In the web service level, the different decision making under uncertainty willresult in different probability of success of the service execution, resulting in thedifferent quality of service (time delay, cost overflow, that need to find the optimaldecision strategy, making the affected area of quality of service as small as possible.Different type of uncertainty events that lead to different type of state, UTG(Uncertainty Triggering Graph) is presented to describe the relations between uncertainstates and decision strategies. Then, Markov Decision Processes (MDP) is proposed tosolve the optimal choice of strategies. Simulation experiments validate the effectivenessof this method, a number of factors affect the decision-making choice are discussed.(2) In the business service level, a periodical multi-instance service process withsynchronization constrains model is proposed, to abstract business services and analyzethe different events corresponding to different actions. Profit and cost of differentactions are calculated. The greedy algorithm is used to obtain the optimal strategy.Similarly, the simulation experiment is given to shows the validity and correctness ofthe method.(3) The simulation environment for decision making under uncertainties for webservice is developed to visualize the uncertainty of the decision-making process, including the planning of the service execution process, uncertain events, serviceexecution status, feasible decision-making action and decision-making results,automatic simulation and manual simulation is executed, to further verify the feasibilityof the method.