Genome analysis in a dynamically scaled hybrid cloud

In this paper, we explore the benefits of automatically determining the degree of parallelism used to perform genetic mutation calling in a hybrid cloud environment. We propose algorithms to automatically control both the hiring of hybrid cloud resources and the selection of the degree of parallelism employed in analysis tasks executed against that cloud. Using the Broad Institute's Genome Analysis Toolkit as a case study, we then conduct profile-driven simulation studies to characterise the circumstances in which our algorithms are beneficial or deleterious compared to simple, conventional baseline algorithms. We find that there are a wide range of cloud workload scenarios where our algorithms outperform the baselines, and thereby argue that automatic control of cloud scaling and task parallelism, using techniques like those proposed, are likely to be beneficially applicable to real-world biocomputing.