Abstract Public cloud storage services are nowadays a data intensive domain and already pro- ducing a dominant share of the Internet traffic world wide. Huge files are synchronized between clients and different data centers of each storage provider. Little is known about the performance of each individual service. Execution times of upload and down- load operations may vary over time. According to current researches the performance depends on daytime and other factors, e.g. euclidean distances to the nearest data center. In this thesis a global distributed cluster of servers will perform periodic interactions with three of the most used public cloud storage provider, i.e. Dropbox, Google Drive, and OneDrive. Collected data is used to obtain insights in the behavior of these ser- vices. Techniques coming from machine learning and statistical analysis are applied in order to infer knowledge about the geographic performance over the daytime for each single provider. Different algorithms are compared with each other, finding the optimal regression model. Performances of each storage is evaluated in order to find the best service in a given situation.