To decide whether the increased overhead of a multi-agent system is necessary to solve the requirements, the problem
should be analysed thoroughly and an informed decision should be made. For this purpose, a number of criteria have been
defined. It is not necessary to answer all of the questions below with the answer that makes a MAS suitable, but if to
many of them can be answered otherwise, the choice for a MAS must be made conciously and with potential drawbacks in
mind.
To decide if implementing an adaptive MAS is useful at the global level, eight criteria are defined. The answers
in brackets are the ones that indicate that a MAS is suitable.
-
Is the global task completely specified? Is an algorithm a priori known? (No)
-
If several components are needed to solve the global task, do they need to act in a certain order? (No)
-
Is the solution generally obtained by repetitive tests, are different attempts required before finding
a solution? (Yes)
-
May the environment of the studied system be dynamic? (Yes)
-
Is the system functionally or physically distributed? Are several physically distributed components
needed to solve the global task? Or is a conceptual distribution needed to solve it? (Yes, Yes, Yes)
-
Is a large number of components needed? (Yes)
-
Is the studied system nonlinear? May a little modification of a local behaviour have a great impact on
the global result? (Yes)
-
Finally, is the system open or evolutionary? Can new components appear or disappear dynamically?
(Yes, Yes)
The local level has to be examined to know whether some components require being recursively designed as adaptive
multi-agent systems. Three criteria were defined to try and answer this question. The answers in brackets are the ones
that indicate that a component can be a single agent instead of a MAS itself.
-
Does a component have a limited rationality only? (No)
-
Is the component “big” or not? Is it able to do many things, to reason? Or does it need simple abilities to perform
its own task? (No, No,Yes)
-
May the behaviour of a component evolve? (No)
This content has been adapted from
Bernon, C., Gleizes, M. P., Migeon, F., & Serugendo, G. D. M. (2011). Engineering Self-organising Systems. In
Self-organising Software, Natural Computing Series (pp. 283-312). Springer Berlin Heidelberg.
|