Deployment of Batch Mode Scientific Workflow on a Computation-as-a-Service Private Cloud

Cloud computing is usually to address business problems of costly computing infrastructures but nowadays it is considered as a possible alternative to scientific workflow deployment. Therefore, there are only limited cases for scientific and engineering computing in which there are task parallelism closely coupled with high concurrent I/O requirements. To address this issue, this paper developed a new resource management methodology to maximize overall machine utilization levels while minimizing application run time. The key strategy and algorithm in this methodology consist of: (i) a bottom-up architecture that utilizes resources for both servers and clients. (ii) a maximum utilization resource coloration algorithm based on node ability. A prototype system was implemented by incorporating the policies and algorithms mentioned above in Cloud Computing and Distributed Systems (CLOUDS) Laboratory. Initial results were obtained by two different cases, by Rotorysics (formerly Propella), a special marine hydrodynamics code for propellers and turbines and by DF_OSFBEM, a panel method code for unsteady 3D multiple-foil hydrodynamics. Results showed that new solution has speeded up total run time up to 50% at the 2nd level - the higher service ability level. By using the developed methodology and exploration of Computation-as-a-Service (CaaS), the objective was achieved to accelerate scientific workflow efficiency in private cloud computing platform.