An Object-Oriented Approach to Large-Scale Battlefield Simulation,
Abstract
Large scale computerized battlefield simulations have been in existence for a long period of time. CEM VI (Concepts Evaluation Model VI), upon which we have directed our effort, was first developed in 1968. Since then, it has evolved through several different authors and types of Fortran implementations. The last critical update occurred in 1983 with the introduction of ATtrition using CALibrated parameters (ATCAL) algorithm. CEM VI is a discrete event simulation. As such, it is subject to random and a priori unknown branching. Thus, data are not contiguous in memory, and the data structure evolves with the simulation. The algorithm, as formulated, was unable not amenable to vectorization on the new Cray architectures. A typical CEM VI simulation, executed in the scalar CPU, typically consumes several to 10 hours of Cray 2 CPU time. To ameliorate this situation, we developed a strategy whereby the kernel of CEM VI (ATCAL) could be vectorized. After careful investigation it was determined that data motion was the key in realizing the potential for vectorizing the ATCAL algorithm. Three different strategies were investigated, with execution rates determined for each method. Taking advantage of the Cray gather/scatter hardware was determined the most feasible of the strategies investigated. After implementing the strategy in ATCAL, a speedup of 8.09 was obtained.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1992
- Accession Number
- ADP006616
Entities
People
- Michael Brewer
- Pat Burns
Organizations
- Colorado State University