Title:  Knowledge-based Management of Legacy Codes for Automated Design"

Author: John Keane (Thesis) - HPCD-TR-49
                         
                            Abstract

Systems for automated design optimization of complex real-world
objects can, in principle, be constructed by combining
domain-independent numerical routines with existing domain-specific
analysis and simulation programs.  Such ``legacy'' analysis codes are
frequently unsuitable for use in automated design.  They may crash for
large classes of input, be locally non-smooth, or be highly sensitive
to control parameters.  To be useful, analysis programs must first be
modified to reduce or eliminate only the undesired behaviors, without
altering the desired computation.  To do this by direct modification
of the programs is labor-intensive, and necessitates costly
re-validation.

This dissertation describes research into how legacy analysis codes
can be usefully employed in design automation systems.  We show that
recovery from failure is possible when the failure occurs in the
context of a search-based process such as optimization.  We discuss
the importance of failure context in determining the correct failure
recovery action.  We then describe an approach to failure recovery
that is both context-sensitive and guarantees the integrity of the
original computation to which it is applied.

We have implemented a high-level language and run-time environment
(together called LCM) that allow context-sensitive
failure-handling strategies to be incorporated into existing Fortran
and C analysis programs while preserving their computational
integrity.  Our approach relies on globally managing the execution of
these programs at the level of discretely callable functions so that
the computation is only affected when problems are detected.  Problem
handling procedures are constructed from a knowledge base of generic
problem management strategies.  We show that our approach is effective
in improving analysis program robustness and design optimization
performance in several real-world design domains.