Research And Implementation Of A Multitask-oriented Cloud-based Service Composition Algorithm

With the rapid development of cloud computing technology and the increasing complexity of application system,a single simple service can not meet the functional requirements in many real situations.Service composition has become an effective solution,which dynamically combines atomic services with different quality attributes to quickly construct large-scale distributed applications.The composite service requests from many concurrent users is rapidly increasing in the background of the cloud computing.Multiple users will look up and invoke the required composite services independently,these requests are then received,queued,and processed by the service composition system.Therefore,it's necessary to efficiently solve the cloud computing service composition(CCSC)problem.There are some simple scheduling algorithms to deal with large numbers of requests for CCSC.However,it ignores the similarities and relationships that may exist in these requests,and does not guarantee the quality of the final solution.So we propose a multitasking algorithm for cloud computing service composition by modeling this problem.On the basis of the multitasking optimization algorithm,named EMA-CCSC,the idea of machine learning is used to group and optimize the requests.Thus the multiple composition service instances are optimized at a time,and the multiple requests can be solved concurrently.The specific contents are as follows:(1)We model the Qo S-aware cloud computing service composition problem,and give its definition.First,we define six quality standards of atomic services in detail,and build the structural model of composite services by using DAG diagram to express the four structures of service composition.Then we give the aggregation rules of service quality attributes and the representation of Qo S value of composite services to build the execution model.Finally,the problem is mathematically modeled and its mathematical expression is given.(2)We design and implement a multitasking optimization algorithm for CCSC problem,called EMA-CCSC.Unlike the existing optimization algorithms,the algorithm can simultaneously optimize two or more instances of CCSC problems.As a result,it can deal with more requests in a given time.The algorithm can promote the convergence of the optimal solution through the transfer of the hidden knowledge,thus it can effectively solve the multiple requests problem while guaranteeing the quality of the solution.(3)We design a suitable group model to group multiple CCSC instances in a request list,then use EMA-CCSC algorithm to optimize these instances.We use EMA-CCSC algorithm to get training samples by doing a lot of experiments on 1188 composite service instances,and then use random forest model to train.The obtained model can predict the gain value of the two instances optimized at the same time,and then we can group these instances based on the predicted value.Thus the instances that can promote each other will be optimized together as much as possible,and the quality of the solution will be better.(4)We carry out the experiments to demonstrate the effectiveness and correctness of the proposed multitasking optimization algorithm.Firstly,we apply EMA-CCSC algorithm to solve 1188 composite service instances base on the QWS data set,then compare them with the optimization results of 9 single task optimization algorithms.Secondly,the instances will be grouped and optimized separately according to the random forest group model and random group model in which the number of atomic services is same,and the optimization results are compared.Finally it is proved that the proposed algorithm is efficient.