Cloud service providers, monitor average resource (for e.g. CPU) consumption and based on predefined limits (for e.g. CPU-Idle-time > 500 milliseconds), provision or de-provision resources. Traditionally this is a reactive approach and doesn't fully address the wide range of enterprise use cases. Implementation of predictive approach to resource management has been rarely reported even though they could perform potentially better than their counterpart. Identification of a suitable model for predicting the performance of the system under a load is an ideal precursor in managing resources on a cloud environment. The current study compares the performance of two such predictive models namely Holt-Winter and ARIMA using a public web server data set Request rate was used as the metric to monitor resource consumption. The experiment results show that Holt-Winter model performs better than a few selected ARIMA models, which could be subsequently used for managing resources on cloud if the data request rates follow a similar pattern. © 2014 IEEE.
|Journal||Data powered by Typeset2014 14th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing|
|Publisher||Data powered by TypesetIEEE|