Experience with KQML

Previous: KQML Software Architectures Up: KQML as an Agent Communication Language Next: Conclusion

Experience with KQML

The KQML language and implementations of the protocol have been used in several prototype and demonstration systems. The applications have ranged from concurrent design and engineering of hardware and software systems, military transportation logistics planning and scheduling, flexible architectures for large-scale heterogeneous information systems, agent-based software integration and cooperative information access planning and retrieval. KQML has the potential to significantly enhance the capabilities and functionality of large-scale integration and interoperability efforts now underway in communication and information technology such as the national information infrastructure and OMG's CORBA, as well as in application areas electronic commerce, health information systems and virtual enterprise integration. The content languages used have included languages intended for knowledge exchange including the Knowledge Interchange Format (KIF) and the Knowledge Representation Specification Language (KRSL) [21] as well as other more traditional languages such as SQL. Early experimentations with KQML began in 1990. The following is a representative selection of applications and experiments developed using KQML.

The design and engineering of complex computer systems, whether exclusively hardware or software systems or both, today involves multiple design and engineering disciplines. Many such systems are developed in fast cycle or concurrent processes which involve the immediate and continual consideration of end-product constraints, e.g., marketability, manufacturing planning, etc. Further, the design, engineering and manufacturing components are also likely to be distributed across organizational and company boundaries. KQML has proved highly effective in the integration of diverse tools and systems enabling new tool interactions and supporting a high-level communication infrastructure reducing integration cost as well as flexible communication across multiple networking systems. The use of KQML in these demonstrations has allowed the integrators to focus on what the integration of design and engineering tools can accomplish and appropriately deemphasized how the tools communicate [10][8][23][17].

We have used KQML as the communication language in several technology integration experiments in the ARPA Rome Lab Planning Initiative. One of these experiments supported an integrated planning and scheduling system for military transportation logistics linking a planning agent (in SIPE [4][30]), with a scheduler (in Common Lisp), a knowledge base (in LOOM [22]), and a case based reasoning tool (in Common Lisp). All of the components integrated were preexisting systems which were not designed to work in a cooperative distributed environment.

In a second experiment, we developed a information agent consisting of CoBASE [6], a cooperative front-end, SIMS [2][1], an information mediator for planning information access, and LIM [26], an information mediator for translating relational data into knowledge structures. CoBASE processes a query, and, if no responses are found relaxes the query based upon approximation operators and domain semantics and executes the query again. CoBASE generates a single knowledge-based query for SIMS which using knowledge of different information sources selects which of several information sources to access, partitions the query and optimizes access. Each of the resulting queries in this experiment is sent to a LIM knowledge server which answers the query by creating objects from tuples in a relational database. A LIM server front-ends each different database. This experiment was run over the internet involving three, geographically dispersed sites.

Agent-Base Software Integration [18] is an effort underway at Stanford University which applying KQML as an integrating framework for general software systems. Using KQML, a federated architecture incorporating a highly sophisticated facilitator is developed which supports an agent-based view of software integration and interoperation [16]. The facilitator in this architecture is an intelligent agent used to process and reason about the content of KQML messages supporting tighter integration of disparate software systems.

We have also successfully used KQML in other smaller demonstrations integrating distributed clients (in C) with mediators which were retrieving data from distributed da. Mediators are not just limited distributed database access. In another demonstration, we experimented with a KQML URL for the World Wide Web. The static nature of links within such hypermedia structures lends itself to be extended with virtual and dynamic links to arbitrary information sources as can be supported easily with KQML.

finin@cmsc