This approach has been successfully applied for vector supercomputers but now the search is on for effective means to make increasing use of parallel supercomputers, particularly methods that will protect the investment made in large, serial codes. Few generally applicable methods for exploiting parallelism in existing codes have been found. Fewer are readily scalable. Parallelisation of "dusty deck" codes often makes disappointingly inefficient use of parallel computing resources.
In parametric studies a range of different simulations are calculated using the same program, each simulation computing output variables for one particular set of input values. Existing techniques for controlling such studies are time-consuming and prone to error. Typically, the use must create a set of input files for each of the parameter settings, run the simulation program against each set of files, collate and merge the output files and produce some condensed form of output.
Parameterised simulations can require enormous amounts of processor time. Even if the evaluation of a single set of parameters does not require a great deal of time (which is rarely the case), the combinatorial explosion met when several variables must be explored can amount to days of serial computation. It is in this context that this paper discusses the use of a tool for managing the execution of parameterised simulations on a distributed memory, parallel supercomputer. A brief description of the tool used is given, and its efficacy illustrated by way of several case studies.