Distributed execution of dynamically defined tasks on microsoft azure

Abstract

Microsoft Azure is a relatively new public cloud service which has potentially much to offer to the scientific community, but has not yet been widely used for research applications. This paper evaluates suitability of Microsoft Azure as a platform for execution of computational applications in three scenarios: we evaluate dynamic horizontal scaling, distributed execution of a CPU-intensive application-POV-Ray raytracer, and distributed execution of a bioinformatics application-Exon-Visualizer. To this end, we created a Distributed Task Library (DTL), due to lack of free, simple solution for distributed execution of dynamically defined tasks in .NET. In conclusion, we show that while dynamic horizontal scaling is quite slow, Microsoft Azure is a worthy platform for computational applications, offering, in conjunction with DTL, an easy way to speed up CPU-intensive, embarrassingly parallel problems.