Research on Parallel Computation Architecture Based on Object Computing Network
In this paper, a multi-body system dynamics problem is considered as a complex system modeling and computation problem. A parallel programming model and its execution environment are designed to reduce model complexity and to improve computational speed. In this architecture, the problem is divided into two levels: 1. Find a parallel mathematical algorithm to describe the behavior of this multi-body system. 2. Build a unified programming language that could leverage many different pieces of computational resources, such as symbolic computing, and numerical libraries. To tackle the first level, Space-Time Finite Element Method (FEM) is applied. The Space-Time FEM formats of Newton-Euler formula is given, as well as its constraint formula. This algorithm has avoided the loop operation during the time field solving. To tackle the second level, a network-based functional programming language - Object Computing Network (OCN) is designed. OCN is inspired by Object-Process Methodology (OPM). In OCN, every computation behavior is treated as a function, which is constructed by a three element set: {Var, Rule. Condition}. Three basic patterns - Branching, Synchronizing, Merging are given in OCN to prove the flexibility of OCN. An communication interface is designed in OCN to connect different programming languages together. Two multi-body system dynamics computing models - Semi-discretization method and Space-Time discretization method - are constructed with OCN, and a significant contrast in task parallelization is shown by different OCN models.