Benchmarking DOUG on the cloud

Large systems of linear equations with sparse matrices arise often in scientific computing problems and engineering tasks. For efficient solution of such problems iterative techniques like preconditioned Krylov subspace methods are used. Domain decomposition preconditioners are good for reducing the number of iterative steps together efficient parallelisation of the problem. While cloud computing infrastructure has become quite attractive also for the HPC community, this paper gives an overview of DOUG (Domain Decomposition on Unstructured Grids) implementation with the focus of important parameters for parallel performance on computer clusters as well as on the SciCloud (Scientific Computing on the Cloud) environment. We describe the used methods and perform a number of tests for benchmarking the application on both environments.