Service oriented architecture for job submission and management on grid computing resources

Running applications on Supercomputers requires users to learn the low level details of how to interact with the resource manager and scheduler, write job submission scripts, submit and monitor jobs, transfer input and output files etc. These mechanisms differ on different machines and the scientist needs to waste precious time browsing websites and reading lengthy documentation. Even with tools such as Globus or Condor-G, submitting jobs requires users to learn the respective job description languages. This adds an unnecessary learning curve specially for users that want to use packages mostly pre-installed on compute resources and are interested only in the results. Our aim is to allow the user to focus on just the science of the simulation and the results and not the details of job submission and management. In order to do this we have designed a Simulation Application Manager(SAM) based on a Service Oriented Architecture(SOA) that abstracts out the details of the underlying execution environment and provides a mechanism to transfer the necessary science parameters from the user to the job. The choice of using an SOA also ensures that user interfaces to various applications can be quickly and easily generated with most of the code being generated automatically by modern day web services tools available in frameworks such as Axis and CXF.