Section: Application Domains

End-User Tools for Computational Science and Engineering

Another Grand Large application domain is Linear Algebra, which is often required to solve Large Scale Computational Science and Engineering applications. Two main approaches are proposed. First, we have to experiment and evaluate several classical stable numerical methods. Second, we have to propose tools to help end-users to develop such methods.

In addition to the classical supercomputing and the GRID computing, the large scale P2P approach proposes new computing facilities for computational scientists and engineers. Thus, it exists many applications which would use such computing facilities for long period of time . During a first period, many applications will be based on large simulations rather than classical implicit numerical methods, which are more difficult to adapt for such large problems and new programming paradigm as they generated linear algebra problems. Then, implicit method would be developed to have more accurate solutions.

Simulations and large implicit methods always have to compute linear algebra routines. So, they were our first targeted numerical methods (we also remark that the powerful worldwide computing facilities are still rated using a linear algebra benchmark http://www.top500.org ). We especially focused on divide-and-conquer and block-based matrix methods to solve dense problems. We have also studied Krylov subspace methods (Lanczos, Arnoldi) and hybrid methods to solve sparse matrix problems. As these applications are utilized for many applications, it is possible to extrapolate the results to different scientific domains.

Many smart tools have to be developed to help the end-user to program such environments, using up-to-date component technologies and languages. At the actual present stage of maturity of this programming paradigm for scientific applications, the main goal is to experiment on large platforms, to evaluate and extrapolate performance, and to propose tools for the end-users; with respect to many parameters and under some specify hypothesis concerning scheduling strategies and multicast speeds [78] . We have to always replace the end-user at the center of this scientific programming. Then, we have to propose a framework to program P2P architectures which completely virtualizes the P2P middleware and the heterogeneous hardware. Our approach is based, on the one hand, on component programming and coordination languages, and on the other hand, to the development of an ASP, which may be dedicated to a targeted scientific domain. The YML framework provides a solution to the first point since it offers the YvetteML workflow language in order to orchestrate components. This is a very intuitive programming approach and it favors the re-usability of optimized and bug-free components. The abstraction of the underlying P2P middleware is also ensured by YML by means of its back-end mechanism. The end-user of YML can submit a computing task to any kind of peer connected to Internet as long as YML has a back-end in charge of the middleware which is running on this peer. Currently, YML has two back-ends for the XtremWeb and OmniRPC middleware. Another one for Condor will be soon available. The second point concerns the integration of SPIN to YML in order to get a complete programming tool which covers all the needs of the client in order to run applications (based on linear algebra methods) over the Internet. Finally, the conclusion of our work would be a P2P scientific programming methodology based on experimentations and evaluation on an actual P2P development environment.