Semantic Logging in a Distributed Multi-Agent System Sorin Ilie, Mihnea Scafes ¸, Costin B˘ adic˘ a University of Craiova, Software Engineering Dept. Bvd.Decebal 107, Craiova, 200440, Romania {sorin.ilie,mihnea.scafes,costin.badica}@software.ucv.ro Thomas Neidhart, Rani Pinchuk Space Applications Services NV Leuvensesteenweg 325, 1932, Zaventem, Belgium {thomas.neidhart,rani.pinchuk}@spaceapplications.com Abstract The paper presents a semantic logging framework which allows structured information logging in an agent-based dis- tributed system for chemical incident response. The logging framework is “semantic” because it allows semantic inter- pretation of logs according to relationships defined between logging events. For example, this approach could help the reconstruction of the order of events that occurred during the response to an incident, thus giving a detailed view of system execution trace, as well as of agents’ decisions taken at various decision points during the incident management workflow. We intend to use semantic logs (i) for helping experts to analyze and explain system actions and thus improving system response to future incidents, as well as (ii) for training stakeholders by setting the system to run replay- like simulations of past incident management workflows. 1. Introduction The work reported in this paper was carried out in the context of the development of an agent-based distributed collaborative system intended to support and help coordi- nation of human experts, population and specialized insti- tutions that are collectively involved in response to disaster situations – the DIADEM project 1 . More specifically, DIA- DEM project targets potential disasters caused by chemical incidents in industrial and urban areas. DIADEM is based on Dynamic Process Integration Framework – DPIF [10] multi- agent platform, which in turn is built on top of Cougaar agent platform [1]. DPIF allows integration of software components of the DIADEM system: gas monitoring and distribution modeling; electronic emergency response guide; multi criteria decision analysis; scenario based reasoning and decision support workbench. Each component applies only local knowledge to achieve its function. Agents in DIADEM support humans that participate in the process of responding to a chemical incident. During the incident response, agents Sorin Ilie, Mihnea Scafes ¸, Costin B˘ adic˘ a, Thomas Neidhart, and Rani Pinchuk were supported by DIADEM project. Diadem project is funded by European Union under Information and Communication Technologies (ICT) theme of the 7th Framework Programme for R&D, ref. no: 224318. 1. DIADEM project – Distributed information acquisition and decision- making for environmental management: http://www.ist-diadem.eu/ connect together through service invocations and create a complex incident management workflow [10], [13]. Usually, after an incident response, experts are interested to analyze the complete system activity during the incident to see how stakeholders handled the various situations that occurred during the incident management process. In par- ticular, they are interested in explaining, at least in some situations, what where the causes of the incident. Therefore, assuming that suitable log information was generated by logging agents’ interactions and major decision points of their reasoning processes, experts will be able to analyze log information in order to reconstruct the workflow exe- cution trace. Difficulties of performing this analysis are: (i) log information is distributed throughout the system, and collection and integration of this information is not trivial; (ii) as DIADEM is a distributed system, the use of local time recorded with the log events is not enough for workflow reconstruction and additional causal and domain-dependent “semantic” information must also be logged. The contribution of this paper is represented by proposal of semantic logging framework which allows structured information logging in agent-based distributed systems for management of the response to an incident. The logging framework is “semantic” because it allows causal interpre- tation of logs according to a caused-by relation between logging events. This approach helps the reconstruction of the order of events happened during the response to an incident, thus giving a detailed view of system execution trace, as well as of agents’ decisions taken at various decision points during the incident management workflow. Semantic logs captured during system execution are intended (i) to help experts analyze and explain system actions through an after- action analysis process and thus to improve system response to future incidents, as well as (ii) to train experts by allowing the system to run replay-like simulations of past incident management workflows. Relevant works addressing distributed logging are [4] and [12]. [4] introduces a framework for distributed logging in JADE multi-agent systems [9]. However, semantic aspects of the logs are not considered. [12] considers only semantic aspects of logging in distributed workflows that address business and inter-organizational needs. The rest of the paper is structured as follows. In Section 2 we define logging principles, such as when and what to – 265 – 978-1-4244-7433-2/10/$26.00 ©2010 IEEE