A Benchmark Model for the Creation of Compute Instance Performance Footprints

Cloud benchmarking has become a hot topic in cloud computing research. The idea to attach performance footprints to compute resources in order to select an appropriate setup for any application is very appealing. Especially in the scientific cloud, a lot of resources can be preserved by using just the right setup instead of needlessly over-provisioned instances. In this paper, we briefly list existing efforts that have been made in this area and explain the need for a generic benchmark model to combine the results found in previous work to reduce the benchmarking effort for new resources and applications. We propose such a model which is build on our previously presented resource and application model and highlight its advantages. We show how the model can be used to store benchmarking data and how the data is linked to the application and the resources. Also, we explain how the data, in combination with an infrastructure as code tool, can be utilized to automatically create and execute any application and any micro benchmark in the cloud with low manual effort. Finally, we present some of the observations we made while benchmarking compute instances at two major cloud providers.