Dynamic Resource Scheduling For Meshed Web Systems In Cloud Computing Environment

The high computing power and the strong robustness of Web applications and Web services(collectively referred to as Web systems)are realized by migrating to Cloud Computing platforms.In order to improve the ability of intelligent services,machine learning,artificial intelligence and other technologies are gradually applied to deal with the increasingly complex business requirements from users,which makes structures of Web systems complex and functions provided by Web systems diverse.And the existing Cloud resource provisioning methods for single-tier and linear multi-tier Web applications cannot supply resources for Meshed Web systems effectively.Thus,it is one of the important issues for Cloud Service providers to design methods to schedule resources elastically and control performance for Meshed Web systems.In addition,due to dynamic fluctuations of time-varying workloads and continuous changes of user demands accessing different tiers,there are inevitable bottlenecks in Meshed Web systems which leads to over-utilization of resources and a sharp drop of system performance.And the complicated transmitting relationships of workloads among tiers also affect the efficiency of bottleneck elimination.Therefore,how to eliminate bottlenecks while avoiding their transferring among tiers becomes another significant problem in elastic resource provisioning for Meshed Web systems.Strategies to provision Virtual Machines(VMs)elastically and eliminate bottlenecks for Meshed Web systems are proposed from the perspective of Cloud Service providers,which are designed to meet constraints of request response time in Service Level Agreement(SLA)at a minimal VM renting cost.They also consider the impacts of many factors such as complex structures,dynamic fluctuations of workloads and the billing methods of resources.The main contributions are:(1)The transferring relationships among tiers in Meshed Web systems can be modeled based on a finite-state discrete Markov stochastic process,and the queuing theory is applied to model the workload processing in each tier.Then the Jackson queuing network can be used to describe the workload behaviors in the whole Meshed Web system accurately.(2)A Jackson queuing network based method to provision VMs and eliminate bottlenecks for Meshed Web systems is proposed,which can be employed to predict the workloads transferring among tiers while eliminating bottlenecks to avoid bottlenecks shifting.And the control theory and the queuing model are combined to design an independent control system for each tier to improve the accuracy of the predicted workloads.(3)State-space model is applied to model the linearized relationships of characteristics in Meshed Web systems in stable state approximately,which can reduce the impacts of the adopted queuing model and the specific workload patterns on system performance.And the system performance of each tier can be controlled jointly by constructing a dynamic close servo system combining state feedback and integral control.All methods for elastic resource provisioning are evaluated based on the simulated Meshed Web system developed on Cloud Sim.The Jackson queuing network based VM scheduling strategy can eliminate bottlenecks while avoiding their shifting among tiers effectively.And the SLA violations can be further reduced by the independent feedback control based method while avoiding bottlenecks shifting.What's more,when dynamic fluctuations of workloads are relatively stable,the SLA violations can be reduced effectively by the State-space model based method,and the mean response time of all workloads in each tier can be stabilized near its corresponding reference value.