Network based parallelism of probabilistic composite structural analysis

Probabilistic structural analysis (PSA) is essential to ascertain the reliability of composite structures to meet the challenge posed by future aeropropulsion systems. The probabilistic method should account for all naturally-occurring uncertainties including those in constituent material properties, fabrication variables, structure geometry and loading conditions. In this paper, parallel computing of probabilistic structural analysis is described and demonstrated to reduce the computational time. The parallelism is accomplished by integrating the computer code IPACS (Integrated Probabilistic Assessment of Composite Structures) with a software system PVM (Parallel Virtual Machine) for heterogeneous concurrent computing in network based computing environments. NASA Lewis's LACE (Lewis Advanced Cluster Environment) computer system which combines 33 IBM Risk/6000 workstations is the testbed for the study and Ethernet system is used for message passing. Two composite structures are analyzed using a probabilistic method with parallel computing for demonstration. It is found that turnaround times of peak-hour runs range from one to two and half times of that of benchmark run in this study. Using 32 workstations for parallel computing, about 90% turnaround time reduction of benchmark run and about 70 to 80% turnaround time reduction of peak-hour cases (comparing with the time of benchmark run using single workstation) can be achieved. It is also found that additional computational time reduction is insignificant (2%) when workstations are increased from seventeen to thirty two. This is because of data transfer congestion. Therefore high speed data transfer system such as FDDI (Fiber Distributed Data Interface) interface (which is ten times faster than Ethernet system) is recommended. Overall, it is concluded that network based parallelism is an efficient and effective approach to speed up the PSA computation.