An Extensible Framework for Predictive Analytics on Cost and Performance in the Cloud

• Published in: 2015 International Conference on Cloud Computing and Big Data (CCBD) pp. 13 - 20
• Main Authors: Sanping Li, Yu Cao, Tao, Simon, Xiaoyan Guo, Zhe Dong, Sun, Ricky
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.11.2015
• Subjects: Analytical models; Application Performance; Cloud computing; Data models; Machine Learning; Monitoring; Predictive Analytics; Predictive models; Real-time systems; Resource Cost; Time series analysis
• DOI: 10.1109/CCBD.2015.23

As we are moving to the cloud, one challenge is the pressure to provide an accurate picture of ongoing resource costs and associated application performance. While cloud offerings give great flexibility to elastic applications, tenants lack guidance for choosing between multiple offerings. The lack of knowledge could lead to tenants over-provisioning and paying for resource that they do not actually need, or under-provisioning with suffering performance issues. In this work, we propose an extensible framework for predictive analytics on cost and performance in the cloud. Resource consumption data is collected and placed at readiness for enabling immediate analysis such as billing with the models of pay-as-you-go and lease. The time series data stored in a tiering object store supporting fast retrieve, as well as the heterogeneous types of data on application events and performance, are utilized to facilitate pattern analysis. These data aggregation, meanwhile, is put into considerations concerning correlation between cost and performance and their changing trends over time. Thus, by leveraging what-if analysis and real-time prediction, the framework gives a quite precise view of current status on cost and performance, as well as future perspectives, so as to support decision making on resource configuration with satisfaction of application's Service Level Agreement (SLA) requirements.