Abstract
Multiagent platforms need to be evaluated focusing on the underlying computer architecture in order to allow developers to exploit the parallelism available in multicore processors. This paper presents the characterization of Jason, a well-known Java-based multiagent platform, when executed on distributed shared memory architectures. Since this kind of architecture is already present in current multicore processors, this should be the first step for the characterization of this platform on distributed systems. To this end, we propose the execution of a set of benchmarks recently proposed for evaluating multiagent platforms. The results obtained show that Jason can be used to program CPU-intensive multiagent applications without loosing the Java scalability over multicore processors. Though, Jason's performance for communication-intensive applications depends on the traffic pattern generated by the agents, the layout of the cores and the selected execution mode (i.e. synchronous or asynchronous).