Normative Control: Controlling Open Distributed Systems with Autonomous Entities

Abstract

Open distributed systems consisting of a potentially large set of autonomous entities might not be controllable directly. More precisely, standard control interventions, such as altering parameters and behaviour, are not possible due to the entity's autonomy. However, indirect control using socio-inspired mechanisms can be applied to guide the system's behaviour and influence the distributed entities using sanctions and incentives. The demanded behaviour as well as the corresponding sanctions and incentives are coded as norms and generated in response to perceived environmental and internal conditions. Such a norm is issued by centralised authorities. Norm violation is monitored using a higher-level observer in a distributed manner. After an introduction and motivation for using social mechanisms in technical systems, we present a novel normative control loop establishing the afore-described concept within a Trusted Desktop Grid scenario. The evaluation demonstrates the potential benefit in terms of an increased system robustness and fast recovery from attack states.

Trust Communities: An Open, Self-Organised Social Infrastructure of Autonomous Agents

“…A depicts the area of the utility deviation, effectively representing the total cost of the disturbance. For further details, see [26]. For A, we do not assume t drop D 0, but we follow the assumption of linear utility change, resulting in Equation (5.5).…”

“…During this phase, the TCM can assign roles to other members in order to distribute the effort of administrating the eTC, it establishes norms to determine the behaviour of all agents inside the eTC, and accepts agents to join or excludes members from the eTC [26]. Members inside the eTC can compare their own predicted utility to the (actual) utility periodically.…”

“…BOINC has at least 50 % overhead, since it replicates every WU at least once (for more details see [26]). A panoramic view on computational trust in multi-agent systems can be found in [27] or [15].…”

To Adapt or Not to Adapt: A Quantification Technique for Measuring an Expected Degree of Self-Adaptation

“…We do not restrict the system composition, i.e., we allow open systems [13]. Each a i can be controlled by different users and it may be designed and developed using arbitrary concepts.…”

The Social Concept of Trust as Enabler for Robustness in Open Self-Organising Systems

“…Game-theoretical considerations show that trust can help to avoid getting trapped in the tragedy of the commons. Kantert et al [12] provide such lines of thoughts in the context of Desktop Grid Computing. In general, trust induces a probability distribution over types of interaction partners of different trustworthiness in a Bayesian game.…”