Design of an international sparse linear algebra expert system relying on an OGF GridRPC data management GridRPC system

The time to solution of linear equation solver strongly depends on the type of the problem, the target computer architecture, the selected algorithm, its implementation and the control parameters. Thus huge work has been performed to propose GridTLSE whose expert web site part provides assistance to a user in choosing the right solver for his problems as well as appropriate values for the control parameters of the selected solver, in addition to a collection of well-known matrices. GridTLSE manages entire expertises for the user, managing data transfers, tasks executions, and graphical charts to help analysing the overall execution. GridTLSE uses the open source GridRPC DIET middleware to distribute computing tasks over the grid platform. Data required by the computational services are matrices whose size can vary from some KBytes to several GBytes, so their management on the platform impacts performances. We present in this paper performance results showing how GridTLSE fulfill its goals, taking care of user good experiences related to time to completion, by using the open source data management middleware DAGDA along with the DIET framework.