Global deterministic and stochastic optimization in a service oriented architecture

Tyler Chang, Layne T. Watson

Abstract

Service ORiented Computing EnviRonment (SORCER) is a Java-based network-centric computing platform. SORCER provides a service oriented architecture, which enables the implementation of parallel algorithms in a dynamic distributed computing environment. SORCER is often used for multidisciplinary aircraft design analysis and optimization. However, the current approach often assigns intense optimization algorithms to run entirely on single overloaded nodes, rather than evenly distributing the workload. The goal of this work is to provide lower-level optimization algorithms as integrated SORCER services and study the overhead of doing so. VTDIRECT95, a Fortran 95 implementation of D. R. Jones' algorithm DIRECT, is a highly parallelizable derivative-free deterministic global optimization algorithm. QNSTOP is a parallel quasi-Newton algorithm for stochastic optimization problems. The potential benefit of integrating VTDIRECT95 and QNSTOP into the SORCER framework is to provide dynamic load balancing among computational resources at the optimization level, resulting in a dynamically scalable process.

Chaitra Raghunath, Tyler H. Chang, Layne T. Watson, Mohamed Jrad, Rakesh K. Kapania, Raymond M. Kolonay:
Global deterministic and stochastic optimization in a service oriented architecture. SpringSim (HPC)2017: 7:1-7:12