AHPCRC Parallel Processing

Parallel Physically Based Object Oriented Simulation


Investigators

S. Sarathy and O. Olatidoye Clark Atlanta University - Army HPC Research Center

Objective

Conventional approaches to modeling complex physical systems are based on functional model paradigm. These methods have proven to be very effective for isolated single use, with a dedicated group of support personnel. However, in today's economic climate, this is an unlikely scenario, instead affordability becomes a key criterion in evaluating the effectiveness of a new product. Modeling environments in particular are driven towards higher degrees of reusability and reconfigurability. Object oriented paradigms have provided the framework to achieve a high level of model and component reuse, together with a component level reconfigurability. Physical Object Oriented Modeling (P/OOM) is a subset of this broader approach, applied to a the domain of vehicle dynamics. P/OOM provides a uniquely transparent and intuitive interface to the model developer, enhancing a direct correspondence with the actual system being modeled. P/OOM also provides to automated or semi-automated methods to parallelize such models, by making use of the intrinsic parallel nature of the model. This effort seeks to apply this technique to the high fidelity modeling of complex ground vehicles. These models will be parallelized for use in real time simulation applications. This effort will focus on algorithms to improve the automated/semi-automated task allocation methods used to develop the parallel versions of these models.

Relevance

The modeling environment developed, can be used to develop families of reusable and reconfigurable models. Incremental validation of model variants together with the simplicity of the configuration management schemes made possible by this approach will significantly lower the cost of maintaining distinct, different and often incompatible, simulation models of Army ground vehicles. In addition the parallel algorithms developed here, will allow affordable simulators to be built around such a high fidelity simulation, using commercial off the shelf parallel platforms.

Methodology

A subset of the object oriented modeling methodology customized to vehicle dynamic physical phenomena (P/OOM) has been developed over the last several years, applied to rotorcraft simulation models. This approach will be now applied to the problem of ground vehicle dynamics. The basic modeling environment is based on representing the physical form and function of the actual vehicle, using a library of extremely abstracted and encapsulated primitive component classes. This model is reduced to a task tree and partitioned automatically using task set algorithms, for MIMD type parallel architectures. Such an approach, though commonly adopted, is inefficient giving rise to extreme sized task sets. Such extremes are usually unacceptable even for shared memory machines. This is partly due to this approach not accounting for synchronization/communication overheads. We are developing a hybrid approach to enhance the efficiency and effectiveness of these task pallcoation methods. Our approach includes the use of clustering algorithms to improve load balancing, as well as allow the user to embed parallel directives within the model to take advantage of topological parallelisms inherent in the problem domain.

1. "Parallel physically based object oriented simulation models", S. Sarathy and O. Olatidoye. Jan 1996, SCS Western MultiConference, San Diego CA.
2. "Physical Object Based Simulation Modelling: Is it real ?", S. Sarathy and P. van Gool. Oct/Nov 1995 CEAS-95 Making It REAL Simulation Symposium, Delft University, Netherlands
3. "Parallel Physical Object Based Simulation : Is it fast and cheap enough ?", S. Sarathy and O. Olatidoye. Oct/Nov 1995, CEAS-95 Making It REAL Simulation Symposium, Delft University, Netherlands