Affiliations: [a] Department of Information Systems, Wrocław University of Science and Technology, Wrocław, Poland
| [b] Nova-System Filip Nowakowski, Wrocław, Poland
Correspondence:
[*]
Corresponding author. Dariusz Król, Department of Information Systems, Wrocław University of Science and Technology, Wyb. Wyspiańskiego 27, 50-370 Wrocław, Poland. Tel.: +48 71 320 3258; Fax: +48 71 320 3884; E-mail: [email protected].
Abstract: This paper assesses the possibility of using a popular middleware platform based on a multiple agent paradigm in full compliance with Real Time Specifications for Java. Two reference scenarios are discussed: one testing thread-to-thread activation (creating and releasing threads), the other featuring agent-to-agent execution for road traffic simulation (creating and releasing agents). These preemptive tasks must be scheduled with minimum delay, therefore, timing correctness as thread-to-thread activation latency and agent-to-agent execution latency is a critical performance index. Given this requirement, the study presented describes an empirical investigation of timing and capacity impact on two platforms: Solaris and Windows in both RT and non-RT versions. The experiments showed the impact on performance of platform type and of capacity with the non-RT approach in particular yielding less accuracy but better stability in agent execution. Suggestions as to how this real-time multi-agent approach might be made more effective are included in the paper.
Real-time (RT) computing and multi-agent systems (MAS) have established their positions, but do they actually fit together? This paper examines if there is any hope of bringing RT into MAS context and if both concepts can exist in the same project.
Why is it important?
This paper discusses the integration of two paradigms to meet soft real-time constraints within a distributed middleware based on both MAS and RT specifications. From a technological point of view of this integration, the findings show that this integration is possible. It demonstrates that results for non-immense multi-agent applications are acceptable in that they exhibit a nearly linear scalability of the platform in the range considered.
