Analysis of Jason’s performance for crowd simulations Víctor Fernández, Francisco Grimaldo, Miguel Lozano, Juan M. Orduña Departament d’Informàtica, Universitat de València, Avda. Vicent Andrés Estellés s/n, 46100, Burjassot, Spain ferbau@alumni.uv.es, francisco.grimaldo@uv.es, miguel.lozano@uv.es, juan.orduna@uv.es Abstract Large-scale crowd simulations require distri- buted computer architectures and efficient pa- rallel techniques to achieve the rendering of visually plausible images while simulating the behaviour of crowds of autonomous agents. The Java-based multiagent platforms, devo- ted to provide the agents with the required lifecycle, represent a key middleware in crowd systems. However, since they are oriented to maximize portability and to reduce the de- velopment cost, they may reduce performan- ce and scalability, two important requirements in large-scale crowd simulation systems. This paper studies the performance and scalability provided by Jason, a well known Java-based BDI-MAS platform, as a plausible framework to be used for large-scale crowd simulations. The performance evaluation results show that some improvements should be performed in or- der to make Jason a suitable middleware for large-scale crowd simulations. 1. Introduction Crowd simulation can be considered as a spe- cial case of Virtual Environments where the avatars are intelligent agents instead of user- driven entities. Each of these agent-based en- tities can have its own goals, knowledge and behavior. In recent years, crowd simulation has become an essential tool for many virtual environment applications in education, mobi- lity, safety and security in public spaces or en- tertainment [8]. However, simulating the rea- listic behavior of large crowds of autonomous agents is still a challenge for several computer research communities [9, 5]. Scalability is a key feature in crowd simula- tion and the simulation of high number of au- tonomous pedestrian-agents represents an ac- tive research area at the intersection of compu- ter graphics, artificial intelligence and distri- bute computing. Any scalable model can be di- vided into three levels. Firstly, graphic engines must render complex crowded scenes as fast as posible. Here, the challenge is to display realis- tic big size crowded scenes at interactive frame rates. However, most of the current (graphics oriented) crowd systems use specific MAS with centralised architectures, so they can hardly si- mulate a few thousands of agents and it is also very diffucult to scale these systems up. Se- condly, the MAS platform represents a midd- leware between the distributed computer ar- chitecture and the graphical engine. The MAS platform mainly addresses two important is- sues, the agents behavior modeling and their parallel lifecycle execution. Some researchers have been testing the performance of existing agent platforms [6], showing a lack of perfor- mance and scalability in many of them. The main challenge that crowd simulation offers to the MAS platforms is the ability of hand- ling a massive and concurrent action proces- sing at interactive rates (i.e. 250ms/action). Thirdly, large scale crowd simulation requires distributed computer architectures (eg. P2P, networked-server,...) to execute the MAS de- signed and to be able to increase the number of agents when required. Hence, scalability is a key issue that mainly depends on the distri- buted computer architecture and the degree of parallelism achieved by the software architec- ture.