Predicting Traffic in the Cloud: A Statistical Approach



Monitoring and managing traffic are vital elements to the operation of a network. Traffic prediction is an essential tool that captures the underlying behavior of a network and can be used, for example, to detect anomalies by defining acceptable data traffic thresholds.
In this context, most current solutions are heavily based on historical time data, which makes it difficult to employ them in a dynamic environment such as cloud computing. We propose a traffic prediction approach based on a statistical model where observations are weighted with a Poisson distribution inside a sliding window. The evaluation of the proposed method is performed by assessing the Normalized Mean Square Error of predicted values over observed values from a real cloud computing dataset, collected by monitoring the utilization of Dropbox. Compared with other predictors, our solution exhibits the strongest correlation level and shows a close match with real observations.


Network traffic analysis, network traffic prediction, sliding window, Poisson process, Dropbox.


Traffic network management in Cloud computing

Related Project

TRONE: Trustworthy and Resilient Operations in a Network Environment (CMU-PT/RNQ/0015/2009)


IEEE International Conference on Cloud and Green Computing 2013, October 2013

PDF File


Cited by

Year 2015 : 2 citations

 B Hong, F Peng, B Deng, Y Hu, et al. DACHmm: detecting anomaly in cloud systems with hidden Markov models, Wiley CONCURRENCY AND COMPUTATION: PRACTICE AND EXPERIENCE, 2015.

 B Hong, F Peng, B Deng, Y Zhang. O-MAP: A per-component online anomaly predicting method for Cloud infrastructure, IEEE International Conference on Information and Automation, Agosto 2015.