F2F-technology Library "JINRLIB" Authors: A.P.Sapozhnikov, T.F.Sapozhnikova You are Language: Fortran, Pascal (C++) Environment: Delphi or C++ Builder visitor here. REENGINEERING TECHNOLOGY FOR AUTOMATION BUILDING OF DISTRIBUTED COMPUTATIONAL SYSTEMS FROM STANDALONE-MADE FORTRAN-WRITTEN PROGRAMS Distributing computational jobs among the Local Area Network is a way of increasing efficiency of large scientific and technical problem solving. The existing standard technologies for programs parallelization, such as MPI or Open MP, are oriented on decomposition onto a number of more small processes. As usual this decomposition is performed manually while software developing or modernization. On the other hand there are no technologies for integration ready-made computational procedures onto more large distributed systems. Our project's main idea is an automation of building computational server from ready-made standalone programs. It permits to integrate well-tested old programs made by previous generation of developers, into modern data-processing systems, including advanced visualization facilities, databases and other ways of human-computer communication. Except that, this approach permits to involve into computational servers developing a lot of experienced experts in numeric methods, not wishing to master new modern tools. That's way the main difference between our approach and traditional point of view on distributed system development is in following: we're proposing a technology of joining instead of technology of decomposition. This is a re-engineering technology, because the main developer's tool became a program for converting of standalone-developed programs onto integrated parts of large distributed computer systems. The general architecture of proposed distributed computational system also seems to be rather non-traditional. User's workstation contains a client-process one of whose aims is to distribute jobs among a number of independent computational servers. Each of them is a separate process, which executes a corresponding program. Requests on input-output operations, needed for server, are interpreted by client. The number of servers and their location on network may be arbitrary, depending of client's requirement. Particularly server can be executed on the same user's workstation. User interaction is performed by client exclusively. Sources and detailed description (in Russian, format .pdf) are submitted. The investigation has been performed at the Laboratory of Information Technologies JINR and supported by the RFBR grant 03-07-90347. |