Agent-based Simulation on a Hybrid, Heterogeneous Computer Cluster: Networked Instruction, Multiple Data (NIMD)

Agent-based Simulation on a Hybrid, Heterogeneous Computer Cluster: Networked Instruction, Multiple Data (NIMD)

The paper and demonstration details a system and method of agent based simulation in contexts to be simulated on a hybrid, heterogeneous compute cluster. Specifically, we shall demonstrate an extension of the Systems Modeling Language (SysML) into temporal and environmental contexts. The resulting models form input, bias and outputs to a topology and weight evolving artificial neural network (TWEANN). The network is evolved on the cluster using a massively parallel implementation of NVidia's Compute Unified Device Architecture (CUDA).

The simulation shall be performed on GenCluster, a hybrid, heterogeneous Rocks/CUDA cluster located in Albuquerque, NM.

http://www.rocksclusters.org/rocks-register/details.php?id=1469

We shall show how these ABM mod/sim/vis techniques may be applied to general problem spaces, but we shall also provide insight into a specific implementation of an intelligent distributor (ID) that is used to partition tasks and data for MPP used within the compute cluster to improve effectiveness and efficiency of the compute fabric.

This novel computational fabric is termed Networked Instruction, Multiple Data (NIMD).

Kenneth A. Lloyd - Watt Systems Technologies Inc. Greg Scantlen - Creative Consultants, LLC. Gary Scnntlen - Creative Consultants, LLC

• Presentation in PDF