Tracking Causality by Visualization of Multi-Agent Interactions Using Causality Graphs
Programming multi-agent systems is a hard task and requires tools to assist in the process of testing, validation and verification of both MAS specifications and source code. In this paper, we propose the use of causality graphs, adapted to the context of
- PDF / 922,001 Bytes
- 15 Pages / 430 x 660 pts Page_size
- 1 Downloads / 169 Views
bstract. Programming multi-agent systems is a hard task and requires tools to assist in the process of testing, validation and verification of both MAS specifications and source code. In this paper, we propose the use of causality graphs, adapted to the context of debugging multi-agents systems, to track causality of events produced in interactions among agents in a group. We believe that simple sequence diagrams are not enough to visually track what are the predecessors or causes of a given new event (i.e. an unexpected message or the observation that a message did not came). We propose this kind of graph as an alternative. We redefine the concept of causality graph for this particular field and propose an algorithm for generation of such a graph.
1
Introduction
Multi-agent systems act in a coordinate fashion to achieve their individual and global goals with sufficient level of guarantee. Coordination is done, most of the time, through complex interactions which involves two or more agents. Programming multi-agent interactions is a delicate task because it is prone to errors due to, most of the times, the lack of tools which assist in the production of a verified and validated design and a correct and even automatic implementation of the interactions. Multi-agent systems interactions are instances of interaction protocols definitions. The definition of an interaction protocol is compound of three different parts. The first one is a specification of the possible sequence of messages exchanged between participants. The second one is the semantics of the performatives. The third one, although this part does not always appears in the definition, refers to the kind of content which could appear in the messages exchanged. We can find many examples of these definitions on FIPA-IEEE specifications1 . In this paper, we rely on the first part of an interaction protocol definition, to assist in tracking the causality of a given communicative act. Causality of a concrete event (i.e. a message exchanged, an undesired result of the interaction studied, an unexpected conversation, etc.), in the context of multi-agent interactions, may be loosely defined as the cause which directly or indirectly leaded to generate the event. What we propose in this paper is the 1
www.fipa.org
M. Dastani et al.(Eds.): ProMAS 2007, LNAI 4908, pp. 190–204, 2008. c Springer-Verlag Berlin Heidelberg 2008
Tracking Causality by Visualization of Multi-Agent Interactions
191
use of causality graphs to track causality messages inside multi-agent conversations. In order to do that, we propose the use of causality graphs adapted to the particularities of multi-agent systems, to follow the thread which starts on the event generation and goes back to the root cause of it. We define our own kind of causality graph, and algorithm for its creation, starting from a logically ordered set of messages exchanged, through logical clocks [11,8,13]. Events, as we consider them, refer only to sending and receiving messages. Only these two kinds of events are obvervable by an external
Data Loading...
