Parallel Design Methods for Optimising Lightweight Structures

The gradient based approach to finding the optimum design of a structure is not naturally suitable for parallelisation by the ‘divide and compute’ strategy, which simultaneously computes different independent designs, used by the random search based approach. This paper describes new workload distribution strategies which are used with an existing fine grain parallel analysis method in order to parallelise individual steps of the gradient based design procedure used in the authors' software viconopt . A simple plane frame program has been adapted to act as the testbed for all the strategies presented. Reasonably high levels of efficiency are reported for a relatively coarse grain Transtech Paramid parallel computer.