A Load Balancing Strategy for Android Cloud Testing Platform

• Main Authors: Hao-Tian Yao, 姚皓天
• Other Authors: Chien-Hung Liu
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/z86zz9

碩士 === 國立臺北科技大學 === 資訊工程系研究所 === 103 === Android Cloud Testing Platform (CTP) can be used to verify if the execution results of an Android application on different devices are all correct. It allows multiple Android devices to perform testing concurrently in order to shorten the overall testing time. To assure that every Android device is available for testing at any time, the CTP currently requires maintaining the same number of virtual machines and devices. However, not all the Android devices are needed in each test. This can make some virtual machines running idle and, hence, results in waste of virtual machine resources. In order to manage virtual machine resources effectively, this paper proposes a load balancing strategy for CTP. The strategy considers the job queue length as the workload of a virtual machine. It can dynamically keep a certain number of virtual machines to provide satisfied services and releases some idle virtual machines to improve the degree of resource utilizations. If too many testing tasks are accumulated in the job queues, the strategy will enable CTP to launch additional virtual machines to speed up job processing, reduce overall system loading, improve system throughput, and shorten users’ waiting time. On the other hand, it will allow CTP to shut down the virtual machines to reduce the waste of resources. Experimental results show that the proposed strategy can keep a suitable number of run-time virtual machines to perform testing and maintain a high degree of resource utilization. When the number of testing jobs is below a certain level, the strategy can also reduce the amount of virtual machines on the premise of minimizing the impact of increasing users’ waiting time.