This work covers database technology for FEA applications. The potential of OR data-base technology in this context has been studied and evaluated by implementing MOS, an FEA application integrated with an OR DBMS. The development of the FEA-MOS prototype system has meant that about 5000 lines of C code, and 5000 lines of AMOSQL and Lisp code, have been implemented together with some additional For-tran code. At the same time a number of ForFor-tran routines have been eliminated from TRINITAS. Certain areas have been studied in more detail than others and this selection has mainly been based on the convenience from an implementation perspective rather than giving priority to specific issues. In areas where additional database facilities need to be implemented or where the implementation has not reached a mature state, the dis-cussion and conclusions are based on simplified implementations.
We have presented an architecture for an FEA application that combines an existing FEA program with an OR DBMS. The present approach provides positive effects on both the external and the internal level. Externally, the mediator approach can support, for example, data exchange and transformation, data and operator sharing, data distri-bution, concurrency control among applications and data sources in an EIS system. By embedding an extensible and main-memory resident OR DBMS in the application, im-mediate access is gained to general database capabilities such as storage management,
data modelling, query language, query processing, and transaction processing. Domain-specific data management can be supplied without sacrificing execution efficiency.
More specifically,
• The FEAMOS approach, an original idea of using main-memory resident, extensi-ble, and OR database technology for FEA, has been introduced.
• An architecture for FEAMOS has been designed that tightly integrates the FEA ap-plication with the embedded DBMS. Architectural considerations, such as the influ-ence of data representations and processing location on the overall efficiency have been discussed.
• The architecture of a mediator-based global EIS system incorporating the FEAMOS system has been outlined. It has been shown how the architecture of the mediator system can support sharing, exchange, and combination of data and processing among EIS applications and data sources.
• The applicability of the domain model concept that provides the application with domain modelling, compilation, and optimization capabilities of an OR query lan-guage has been demonstrated. The benefits of query lanlan-guage modelling and ad hoc queries have been shown for various FEA activities. Additional benefits of using domain models that include easier access through a query language, better data de-scriptions (such as schemas), and ad hoc query processing have been presented.
Hence, it is not necessary to re-implement low-level dedicated data structures such as indexes for each new system. Such a reimplementation not only duplicates im-plementation efforts but, as our example shows, may prove less efficient than the highly optimized data management provided by an embedded DBMS. Domain models can also form a base for mediation of domain information among applica-tions and data sources in an EIS system. For example, by providing access to other databases, such as relational DBMSs described in [8] and [123], from the domain model it is possible to build models and ad hoc queries that combine data from other databases.
• AMOS has been extended with foreign data sources for numerical matrix algebra and basic array representations. Specific capabilities of AMOSQL to handle over-loaded and multi-directional foreign functions provide a convenient mechanism for expressing and implementing domain-specific operations as found in numerical ma-trix algebra. This technique provides a more powerful capability of abstraction and reuse in comparison to pure object-orientation. The matrix data source takes advan-tage of these capabilities for implementing and using type-based function dispatch-ing for selectdispatch-ing appropriate matrix operations. An extension of these facilities to include domain-specific optimization techniques, including dynamic query optimi-zation, to optimize the execution of matrix expressions has also been outlined.
• It has further been shown that the FEAMOS approach provides high processing ef-ficiency for application-critical operations, due to the availability of well-fitted and
tailored main-memory data representations and operations within AMOS. Process-ing efficiency is, however, further dependent on such issues as query optimization, data indexing and filtering, and an appropriate selection of data representation and processing location.
• A number of additional database technologies have been identified as being impor-tant for supporting efficient domain modelling and are acknowledged as potential topics for future research as described in Section 7.2.
The key database technologies that form the basis for this work are object-relational, extensible and main-memory database technologies. The extensibility should further cover the query language, the query processor, and the storage manager:
• The DBMS is object-relational due to the modelling and extensibility capabilities of the query language including overloaded and multi-directional functions.
• Extensibility of the query language provides domain-specific modelling of data and operations. The query processor also needs to be extensible to be able to cope with optimization and cost models for domain-specific foreign operations. Furthermore, the storage manager must be extensible to allow for incorporating tailored data rep-resentations, such as specialised matrix schemas.
• Main-memory is a prerequisite for being able to use the embedded DBMS approach and for supporting high processing efficiency.
Performance measures and comparisons between the original TRINITAS system and the integrated FEAMOS system show that the integrated system can provide competi-tive performance. The added DBMS functionality can be supplied without any major performance loss. In fact, under certain conditions the integrated system outperforms the original system and in general the DBMS provides better scaling performance.
However, a full exploitation of the potential of the DBMS can only be accomplished by intelligent and careful design that takes advantage of DBMS facilities for filtering and accessing data in application operations.
It is argued here that the suggested architecture can form a promising alternative for the design and implementation of FEA applications and similar scientific and engineering applications. The present approach provides positive effects on both the external and the internal level. It facilitates integration of, or communication with, an engineering application with other parts of a global EIS system. By embedding a lightweight and extensible MM DBMS in an application you get a standardized query language for rep-resenting, managing, and exchanging domain data as well as access to generic software facilities for implementation of engineering applications without sacrificing execution efficiency. It will then be possible to achieve a global improvement in the efficiency of FEA software from the point of view of the developer, the maintainer, and the user. It might further be expected that this will result in the increased life-time of data and soft-ware and that the analysis quality can be enhanced.
Generally, database technology can play a similar and important role in the
implemen-tation of scientific and engineering applications of tomorrow, as it is currently doing in administrative applications. Specifically, we believe that database technology adapta-ble to the requirements of engineering applications such as OR DBMSs will play an im-portant role in this area.