A Parallel Multi-agent Spatial Simulation Environment for Cluster Systems

Abstract

For more than the last 20 decades, multi-agent simulations have been highlighted to model mega-scale social or biological agents and to simulate their emergent collective behavior that may be difficult only with mathematical and macroscopic approaches. A successful key for simulating megascale agents is to speed up the execution with parallelization. Although many parallelization attempts have been made to multiagent simulations, most work has been done on shared-memory programming environments such as OpenMP, CUDA, and Global Array, or still has left several programming problems specific to distributed-memory systems, such as machine unawareness, ghost space management, and cross-processor agent management (including migration, propagation, and termination). To address these parallelization challenges, we have been developing MASS, a new parallel-computing library for multi-agent and spatial simulation over a cluster of computing nodes. MASS composes a user application of distributed arrays and multi-agents, each representing an individual simulation place or an active entity. All computation is enclosed in each array element or agent; all communication is scheduled as periodic data exchanges among those entities, using machine-independent identifiers; and agents migrate to a remote array element for rendezvousing with each other. This paper presents the programming model, implementation, and evaluation of the MASS library.

Data Provenance for Agent-Based Models in a Distributed Memory

“…LiSimilar toke Places, Agents can be instructed to invoke a procedure that models an activity. Distributed coordination features such as termination, propagation, and migration are also provided at the simulation logic level [24]. Meanwhile, the mechanisms required to orchestrate such distributed manipulation are handled within the framework.…”

“…Two simple simulations illustrate the basic requirements for use case support while providing a basis for performance overhead measurement: SugarScape [26] and RandomWalk [24]. While these simulations do not serve to illustrate the capability to scale to complex reasoning, they help to form a baseline for model understanding and to show challenges imposed by the processing environment.…”

“…Primarily a test simulation, RandomWalk offers the ability to analyze simple properties of the physical network. In the original simulation [24], agents migrate to one of four randomly selected neighboring locations. A modified version of RandomWalk is used to provide some variety in network communication while evaluating scalable capture of framework operations.…”

Network Motif Analysis in Clouds - Subgraph Enumeration with Iterative Hadoop MapReduce

“…From [10] In non-MapReduce based approaches, Kipps et al [10] evaluated the MASS library [9] against MPI based approach. They showed that MASS places-based implementation as well as MPI-based implementation, see Figures 6 and 7, showed good parallelization characteristics by improving runtime with increasing number of available computing cores.…”

“…Apart from MapReduce, there are other parallelization techniques such as MPI/OpenMP [8] and MASS [9] which overlays data in a distributed array through which a program can access on different nodes. Along with distributed multidimensional array (which are called Places in MASS), MASS library also provides a concept of agents which are set of execution instances that can reside at a single place, access its public data/method members, migrate to other places, spawn child agents, and interact with other agents.…”