Available Bandwidth Estimation In Public Clouds

Recent years have witnessed a rapid increase of public cloud adoption as a result of the growing user demands and numerous advantages of public cloud computing. However, the great benefits of public cloud computing come at the cost of little visibility of the underlying network resources. Motivated by the strong desire of users to gain more information of the network resources, in this paper we experimentally study whether the current Available Bandwidth Estimation Tools (ABETs) can accurately estimate the available bandwidth of public clouds. We find that the current ABETs do not work well in public clouds, and their inaccuracy is caused mainly by the two fundamental virtualization techniques, computer virtualization and network virtualization. Our study has also provided us with important insights into improving the accuracy of the current ABETs or designing new ABETs on public clouds.