Resource Demand Forecasting Approach Based on Generic Cloud Workload Model

摘要

Ubiquitous computing environment generates massive amounts of data exploiting any kind of device, in any location and in any format. IaaS resources which are distributed around the world and can be rented on demand become popular to store and process those data. Therefore, it becomes imperative that an effective methodology is developed to learn what IaaS resources are required, how many resources and for how long are needed. However, the heterogeneity and hybridity of data and applications, and the variation of workload make resource forecasting be a big challenge. In this paper, we purpose a unified resource demand forecasting approach based on the generic application model and generic workload variation model. The approach suits for different hybrid applications, various resources and diverse time-varying workload patterns. Taking input from parameterized cloud applications, workload variation model and resource profile, the corresponding resources demands during any time interval can be deduced as output. Experiments are conducted taking MapReduce application WordCount as the example. Our approach is evaluated by contrasting the logged data against forecasting data, and our results show that the average forecast accuracy rate can reach 90%.