An Image Processing-based Test Bench for Performance Evaluation in Hybrid Clouds

Hybrid clouds are combinations of private, community and/or public clouds, and allow the creation of "cloud bursting" deployment models, in which part of the data processing is done in a private cloud and resources from public clouds are used when needed. This model is particularly interesting for CPU-intensive scientific data processing. In order to help determine in which conditions part of the whole of a large data processing task must be sent to a public or private cloud, dynamic benchmarks are required. In this paper we propose an image processing-based test bench that monitors the performance of the clouds resources to gather useful statistics and to determine which resource should be used for a specific task, in order to reduce costs and running time.