AUTOMATIC DERIVATION OF PERFORMANCE CONSTRAINTS AND GENERATION OF DESIGN GUIDELINES

ONE OF THE MAJOR ISSUES THAT EXISTS IN THE DESIGN PROCESS FOR HIGH PERFORMANCE COMPUTERS IS THAT OF DETERMINING, SPECIFICALLY, THE REASONS WHY A PARTICULAR DESIGN ACHIEVES A PARTICULAR PERFORMANCE MEASURE FOR AN APPLICATION PROBLEM OF INTEREST (E.G., A BENCHMARK). THAT IS, IT IS OF INTEREST TO KNOW WHICH CHARACTERISTICS OF THE DESIGN HAVE CONSTRAINED ITS PERFORMANCE AND WHICH CHARACTERISTICS MAKE IT MORE EFFECTIVE THAN OTHER DESIGNS. IF ONE COULD AUTOMATICALLY DERIVE THIS TYPE OF INFORMATION AND PROVIDE IT TO DESIGNERS EARLY IN THE DESIGN PROCESS, THEN THE OVERALL QUALITY OF DESIGNS SHOULD IMPROVE AND THE PROBABILITY OF MISSION SUCCESS SHOULD INCREASE MEASUREABLY. THE PROPOSED APPROACH TO THIS PROBLEM IS TO UTILIZE THE JRS INTEGRATED DESIGN AUTOMATION SYSTEM AS A STARTING POINT AND TO THEN ADD TO IT THE CAPABILITY TO DERIVE THE INFORMATION OF INTEREST. THE PRIMARY EXTENSIONS TO THE SYSTEM WOULD BE IN THE FORM OF COMPREHENSIVE INSTRUMENTATION PACKAGES THAT WOULD BE INTEGRATED WITH THE EXISTING MICROCODE OPTIMIZATION SOFTWARE ELEMENTS. THE INFORMATION OF INTEREST TO A DESIGNER IS OBTAINABLE FROM THESE OPTIMIZATION ELEMENTS IN A FORM THAT CAN BE PROCESSED FOR MEANINGFUL PRESENTATION TO A DESIGNER. ADDITIONAL EXTENSIONS TO THE EXISTING SYSTEM WOULD TAKE THE FORM OF CONTROL ELEMENTS TO RESPOND TO DESIGNERS REQUESTS WHICH RESULT FROM AN EXAMINATION OF THE DATA PRESENTED. THE PHASE I EFFORT WILL FOCUS ON THE SYSTEM DESIGN AND SPECIFICATION ACTIVITIES INVOLVED. A SUITABLE DEMONSTRATION OF THE APPROACH, THE DATA SETS INVOLVED, AND THE EXPECTED USER--SYSTEM INTERACTION WILL BE CONSTRUCTED.