Research On Service Provisioning Problem For Multi-Tier Applications In Cloud

Cloud computing with its highly scalable, highly reliable, pay-as-you-go and other characteristics, has been widely accepted in industry. More and more large-scale network applications migrate to the cloud as a service for people. Web applications that delivered to the cloud can be adaptive scalable as needed to ensure the system performance while greatly reduce the cost and improve the resource utilization. The cloud service provisioning is the key technology in the delivery of web applications to the cloud, the so-called cloud service provisioning is the independent software to deliver a good application to the cloud and according to the application and performance requirements, PaaS allocates the resources available to the application on demand while maintaining the system at run-time.The web applications are generally multi-tier, such as e-commerce applications, social networking applications. Multi-tier application means that the application is divided into multi-tiers, for example Web tier, application tier and the database tier. Compared to the general service provisioning, the service provisioning problem of multi-tier applications in the cloud becomes more complex, the traditional cloud resource allocation methods is no longer applicable and face many challenges:1. The dependency complexity and service feature difference complexity. Dependency complexity of multi-tier applications refers to the interaction between the tiers of the multi-tier applications. On the one hand is to impact workload of each tier, on the other hand is to affect the workload arrival rates of the tiers in the multi-tier applications. Workloads in different tiers become more complex and difficult to predict. Multi-tier applications service features difference complexity refers to the difference of the multi-tier applications hosted services, different service time.2. The capacity of cloud resources complexity and the combination complexity of to multi-class multi-tier hybrid resources. Multi-tier application dependency between the tiers of complexity and service characteristics of resources due to the complexity of the complexity of the processing capacity, i.e. the number of workload distribution and service features within the application request can be effectively related and become very complex; combination of complex multi-tier hybrid multi-class resources refers to different resources available to different tiers of processing power for multi-tier service provisioning. There are a variety of resources for a variety of combinations. How to choose an appropriate mix of resources combination, making the SLA to meet user requirements, the quality of services and resources to achieve a balanced consideration of two contradictory goals, is a technical problem.In this paper, multi-tier applications service provisioning problem in cloud faces challenges, and we mainly research:1. Build online monitoring framework for multi-tier application to monitor workload distribution and make the application load forecasting method based on regression model for solving the complex dependencies between the tiers of the multi-tier application problems; Through monitoring each service features solves the difference between the multi-tier application tiers complexity.2. Use the application of queuing theory to deploy resources for modeling multi-tier application tiers. The resource processing capabilities of each tier is different; proposed performance multi-objective optimization algorithm balanced consideration, pareto optimal application ideas, quality of service and resources to obtain optimum service delivery costs solving the multi-class combination of multi-tier complexity of the problem of mixed resources.In this paper, we use the multi-tier application benchmark RUBiS to do experiments, the proposed method is validated by a large number of experimental data. Based on data collected RUBiS running on workload forecasts and actual operating data and prediction methods in this paper compares the predicted workload. Experimental results show that the workload prediction method in this paper is effective. Prediction method has better performance. On the other hand, we provide the comparative experiment between our strategy and other strategies on resource cost and service overall performance. Experimental results show that compared with similar service provisioning strategy, our strategy can provide the performance-cost equilibrium. That has better performance, quality of service. Our study can improve the efficiency and accuracy of service delivery methods in cloud with high practical value and broad application prospects.