A Resource Management System Of Cloud Computing Service Platform

As a newly emerging technology, Cloud Computing makes IT infrastructure, platform and software as a service through network, and provides on-demand services. With the adoption of virtualization, Cloud Computing can build up a unified resource pool with computing, storage, networking, thus reducing the difficulty of resource allocation and management effectively. How to manage the resource pool effectively to achieve the dynamic and scalability features of Cloud Computing is quite important. Currently, Cloud Computing Service Platforms like OpenStack and CloudStack rely on manual configuration to achieve resource management. Due to cyclical changes of the scale of resources needed and the load of platform, users can’t adjust their resources in time, thus resulting in issues like waste of resources and abnormal execution of services. To solve these problems, this thesis proposes an adaptive resource management system based on OpenStack. The system makes intelligent analysis and decision based on monitoring data of resources, and adjusts the resources of virtual machines dynamically.The main work of this thesis is divided into four parts:Firstly, we design an adaptive resource management framework using DSA (Discovery-Strategy-Action). The framework consists of resource discovery, resource scheduling and resource adjustment. Resource discovery is in charge of detecting status of resource pool and accepting requests from users. Then, resource scheduling generates instructions for adjusting resource. The resource adjustment is responsible for executing these instructions. These three parts work cooperatively to achieve real-time response to resource requirements. Secondly, we design and implement the resource monitoring subsystem based on C/S architecture. The subsystem includes a client module and a server module. The client module finishes the periodic collection of performance data like CPU utilization of physical hosts and virtual machines. The server module is responsible for aggregating monitoring data from clients, persisting them into database, and triggering alarms when resource threshold cross-border is detected.Then, we design and implement the resource decision-making subsystem. The subsystem is responsible for handling alarms of resource load and requests of creating virtual machines. For alarms, the subsystem automatically generates reasonable adjustment decisions according to the statistics of monitoring data and takes the decisions into effect by invoking the dynamic resizing and live migration interfaces of OpenStack. In this way, it can effectively reduce unnecessary adjustments caused by the instantaneous peak. As for requests of virtual machine creation, the subsystem selects a host to place the new virtual machine by using the greedy algorithm. The subsystem can meet resource needs of service dynamically as well as maximize the resource utilization of platform.In addition, we expose the functions of resource monitoring and decision-making subsystems in REST API. Based on these APIs and service interfaces of OpenStack, we design and implement a web console of our Cloud Computing Service Platform. It supports users to customize functions like resource monitoring or resource auto-scale of a specific virtual machine. It also provides visualization of monitoring data, virtual machine management, etc.At last, we conduct some experiments and tests on our system. We also deploy the Multimedia Conferencing System of our lab on the platform to verify the effectiveness of our system. The results show that the resource management system can adapt to the changing resource needs and load timely, and achieve the desired functional requirements.