7 - Indira Gandhi Centre for Atomic Research

5. Another important component of agent scheduling is the communication protocols
among agents. In order to achieve this coordination, the agents might have to
interact and exchange information; therefore they need to communicate by sending
messages. KQML (Knowledge Query and Manipulation Language) is a good
example of communication language.
3. Coordination Models
The design and management of a multi-agent system may take advantage from the
choice and exploitation of a coordination model, that is, a high-level interaction
abstraction aimed at globally ruling the behavior of the different system components.
As said earlier. a coordination model provides a formal framework in which the
interaction of software agents can be expressed and their activities, distribution,
synchronization etc. is maintained. It is comprised of three elements viz., coordinables,
coordination media and coordination laws. The coordinables are the components of the
multi agent systems, i.e. they are the agents themselves, whereas coordination media
provides abstraction for interactions among coordinables and control them. The
coordination laws define how the events are handled when agents interact. The laws can
be defined in terms of a communication language, that is a syntax used to express and
exchange data structures, and a coordination language, that is a set of interaction
primitives and their semantics. Coordination models can be classified as control-driven or
data driven ones.
3.1. Control-driven Coordination Model
In a control-driven coordination model agents typically interact with the external world by
generating and handling events on well defined input output ports. These interactions are
taken care by coordination media. The coordination laws establish how events and state
changes can occur and how they should propagate. Therefore, the coordination media
handle the topology of the interaction space among agents, without paying any attention to
the data possibly exchanged between processes. From the viewpoint of the coordination
media, the only data of interest are the communication events.
3.2. Data-driven Coordination Model
In data-driven coordination models coordinables interact with the external world by
exchanging data structures through the coordination media. These data structures basically
store data, which is shared among different agents. The coordination laws establish how
data structures should be represented and how they should be stored and extracted from
the data spaces. Unlike control-driven coordination models, the coordination media has no
perception of state changes of the coordinables and does not provide for any virtual
connection among coordinables. Different application contexts exhibit different needs
with respect to coordination, and the choice of a coordination model is likely to have a
great impact in the design of multi-component applications. In general, control-driven
coordination models suits better those systems made up of a well-defined number of
entities in which the flow of control and the dependencies between the components have
to be regulated, and in which the data exchanged in not so important. These include, for
example, computational intensive parallel applications, distributed management systems,
definition of complex software architectures. Instead, data-driven model, seems to better
213