AUTOMATED ACQUISITION AND DISSEMINATION OF DISTRIBUTED SYSTEM SOFTWARE DESIGN AND KNOWLEDGE

A NEW GENERATION OF COMPUTERS WILL BE REQUIRED TO SUPPORT THE AIR FORCE IN THE 1990'S AND BEYOND. APPLICATIONS SUCH AS INTEGRATED BATTLE MANAGEMENT, REMOTE SENSING, IMAGE PROCESSING, INTELLIGENT ROBOTICS, AND SPACE VEHICLE CONTROL WILL REQUIRE RELIABLE COMPUTING POWER WHICH EXCEEDS THAT PROJECTED FOR MONOLITHIC COMPUTING ARCHITECTURES. THE COMPUTING DEMANDS OF ADVANCE APPLICATIONS WILL REQUIRE THE EFFECTIVE USE OF DISTRIBUTED PROCESSING SYSTEMS. AS THE NEED INCREASES, THE PERFORMANCE REQUIREMENTS FOR DISTRIBUTED SYSTEM SOFTWARE DESIGN (DSSD) BECOME MORE DEMANDING. IN CURRENT DSSD PRACTICE, PERFORMANCE CONDIERATIONS ARE EITHER IGNORED OR MISAPPLIED. THE RESULT IS SYSTEMS WHICH DO NOT MEET REQUIRED PERFORMANCE STANDARDS AND MUST UNDERGO COSTLY REDESIGN. THIS IS BECAUSE THE EXPERTISE REQUIRED TO MAKE PERFORMANCE TRADEOFFS IN DSSD IS BEYOND THE CAPABILITY OF MOST SOFTWARE DESIGNERS. THE USE OF EXPERT SYSTEMS WITH APPLICATION SPECIFIC DSSD KNOWLEDGE BASES IS A PREREQUISITE TO EFFECTIVE USE OF ADVANCED, HIGH PERFORMANCE DISTRIBUTED COMPUTING ARCHITECTURES. SUCH EXPERT SYSTEMS MUST BE ABLE TO CONSTRUCT LARGE PORTIONS OF THEIR OWN KNOWLEDGE BASES. THE TECHNIQUE PROPOSED FOR KNOWLEDGE ACQUISITION IS TO TAKE AN INITIAL SET OF GENERAL RULES PROPOSED BY HUMAN EXPERTS; A SET OF DESIGN CRITERIA; A DESCRIPTION OF A TARGET DISTRIBUTED SYSTEM; AND USE AN INDUCTIVE LEARNING ALGORITHM TO CONSTRUCT A DOMAIN SPECIFIC SOFTWARE DESIGN KNOWLEDGE BASE.