[ Top ] [ Previous Section ] [ Next Section ]

A Heterogeneous Reasoning and Mediator System

4. Conclusions

Today's world contains a vast quantity of electronic data contained in databases and data structures located across the world. Often, an individual or organization has needs that involve accessing existing, legacy databases, that were created by a different organization for (possibly) an entirely different purpose. In other applications, the very type of data being stored (e.g. pictures, audio, etc.) may necessitate storage techniques that are unique to that form of data. When performing complex reasoning tasks (such as the terrain reasoning example, and the law enforcement example), we may need to access multiple representations of data. More importantly, complex reasoning tasks require that we not only access multiple types of data, but that we access implementations of multiple reasoning paradigms (e.g. face recognition, pattern recognition, etc.).

In this paper, we have used the notion of a ``mediator,'' due to Wiederhold [24,25] as the starting point for our work on integrating multiple databases and multiple software packages. Based on this idea, we have proposed a logic-based language for expressing these mediators in. The language is simple and easy to use and provides a uniform way of passing different types of data structures ``back and forth'' between the mediator and individual software packages. This language for expressing mediators in represents the backbone of our HERMES (``HEterogeneous Reasoning and MEdiator System'') system. We have described the overall systems architecture of HERMES, and also described, in further detail, the different components of the system. Mediators may now be developed within the HERMES framework -- HERMES represents not only a language for expressing mediators in, bu a complete environment for developing integrated mediated systems in. At the current point, our HERMES system inter-operates between a number of different software packages developed commercially (e.g. INGRES, PARADOX, DBASE, ObjectStore in progress), as well as two implementations of reasoning paradigms developed by third parties (face recognition, path planning), as well as some locally developed software packages (e.g. spatial databases, text databases, multimedia databases). HERMES has various B-tree based indexing schemes that allow it to efficiently access over 10 Gigabytes of data on auxiliary storage.

The great challenge for scaling up systems like HERMES is to move from the Gigabyte domain to the Terabyte domain. We are currently investigating techniques for the efficient implementation of a distributed version of HERMES. This would recognize the practical reality that usually, the databases and/or software packages that we may be inter-operating between, are located across multiple nodes on the network. In addition, by utilizing distributed processors, it is hoped that the Terabyte challenge can be addressed. However, at this stage, it is too early to pass judgement on distributed scalability of HERMES.

[ Top ] [ Previous Section ] [ Next Section ]

Click here to go back to the Hermes homepage .

Web Accessibility