Self-organising resource allocation in open systems : theory and experiments

Julia Gertrud Schaumeier, Moez Draief, Jeremy Pitt, UK-India Education And Research Initiative : Qatar National Research Fund
Open, embedded multi-agent systems have applications in sensor and opportunistic networks, and in cloud and grid computing. Features of such systems include: no centralised control, competition for resources between autonomous agents, (un)intentional errors, and a speed and complexity of decisions beyond human capabilities. Moreover, there is a requirement to optimise performance with respect to multiple, possibly conflicting, criteria, for example longevity, occupancy and fairness. This thesis
more » ... addresses the problem of engineering self-organised resource-allocation management schemes for open, embedded systems. Based on the theory of institutions for collective action as defined by political economist Elinor Ostrom, we define a formal model for self-organised resource-allocation, using the computational framework of dynamic norm-governed multi-agent systems. Our model of an electronic institution encapsulates a first-order logic axiomatisation of the principles for enduring institutions. An experimental platform for an abstract common-pool resource management situation has been developed, and the experiments show the importance of all principles in order to achieve longevity, appropriate behaviour and the right balance of membership. The results furthermore suggest that the mechanisms to design institutional rules should be made available to the system components themselves. In order to successfully self-organise, the system has to be aware of its internal state and externalities. To represent and reason about awareness, some aspects from the field of organisational justice have been formalised in the same framework. Agents will not only follow a collectively decided allocation procedure but will execute the allocation according to their own notion of fairness, and also use this notion to judge the perceived behaviour of others. Further experiments show that the ability for introspection and reflection on the perceived environment leads to an improved management profile and further enhances the system's performan [...]
doi:10.25560/14302 fatcat:xmf45macm5dl5aufpd2bfsjmwu