Autonomic Management And Performance Optimization For Cloud Computing Services

Cloud computing has become an increasingly important computing paradigm. It offers three levels of on-demand services to cloud users: software as a service (SaaS), platform as a service (PaaS), and infrastructure as a service (IaaS) . The success of cloud services heavily depends on the effectiveness of cloud management strategies. In this dissertation work, we aim to design and implement an automatic cloud management system to improve application performance, increase platform efficiency and optimize resource allocation. For large-scale multi-component applications, especially web-based cloud applica- tions, parameter setting is crucial to the service availability and quality. The increas- ing system complexity requires an automatic and efficient application configuration strategy. To improve the quality of application services, we propose a reinforcement learning(RL)-based autonomic configuration framework. It is able to adapt appli- cation parameter settings not only to the variations in workload, but also to the change of virtual resource allocation. The RL approach is enhanced with an efficient initialization policy to reduce the learning time for online decision. Experiments on Xen-based virtual cluster with TPC-W benchmarks show that the framework can drive applications into a optimal configuration in less than 25 iterations. For cloud platform service, one of the key challenges is to efficiently adapt the offered platforms to the virtualized environment, meanwhile maintaining their service features. MapReduce has become an important distributed parallel programming paradigm. Offering MapReduce cloud service presents an attractive usage model for enterprises. In a virtual MapReduce cluster, the interference between virtual machines (VMs) causes performance degradation of map and reduce tasks and renders existing data locality-aware task scheduling policy, like delay scheduling, no longer effective. On the other hand, virtualization offers an extra opportunity of data locality for co-hosted VMs. To address these issues, we present a task scheduling strategy to mitigate interference and meanwhile preserving task data locality for MapReduce applications. The strategy includes an interference-aware scheduling policy, based on a task performance prediction model, and an adaptive delay scheduling algorithm for data locality improvement. Experimental results on a 72-node Xen-based virtual cluster show that the scheduler is able to achieve a speedup of 1.5 to 6.5 times for individual jobs and yield an improvement of up to 1.9 times in system throughput in comparison with four other MapReduce schedulers. Cloud computing has a key requirement for resource configuration in a real-time manner. In such virtualized environments, both virtual machines (VMs) and hosted applications need to be configured on-the fly to adapt to system dynamics. The in- terplay between the layers of VMs and applications further complicates the problem of cloud configuration. Independent tuning of each aspect may not lead to optimal system wide performance. In this work, we propose a framework for coordinated configuration of VMs and resident applications. At the heart of the framework is a model-free hybrid reinforcement learning (RL) approach, which combines the advan- tages of Simplex method and RL method and is further enhanced by the use of system knowledge guided exploration policies. Experimental results on Xen based virtualized environments with TPC-W and TPC-C benchmarks demonstrate that the framework is able to drive a virtual server cluster into an optimal or near-optimal configuration state on the fly, in response to the change of workload. It improves the systems throughput by more than 30% over independent tuning strategies. In comparison with the coordinated tuning strategies based on basic RL or Simplex algorithm, the hybrid RL algorithm gains 25% to 40% throughput improvement