Parallel Processing Approaches for Multi Disciplinary Optimization Algorithms

Florin B. Manolache
Dept. of Mathematical Sciences
Carnegie Mellon University
Pittsburgh, PA 15213, USA

Sorin Costiner
United Technologies Co.
UTRC, 411 Silver Ln.
East Hartford, CT 06108, USA

ABSTRACT: Multi Disciplinary Optimization (MDO) problems are often encountered in many industrial areas. MDO refers to the optimization of systems of subsystems (disciplines). The optimization of the full system is often reduced to a hierarchy of optimizations at subsystem and system levels. This paper presents a concise survey of major MDO approaches and discusses opportunities of parallel processing (PP) at four algorithmic levels of the approaches, i.e., the subsystem solver level, and the sequence of problems level. Advantages of different PP implementations of the MDO approaches are outlined. Special emphasis is put on vertical PP processing, where one thread treats hierarchical structure (e.g., a full system evaluation), inter-thread communication is low, and processor loads are uniform.

Get the paper in its entirety as