The Decentralised Coordination of Self-Adaptive Components for Autonomic Distributed Systems

ion Control signal system model Disturbance Inputs Inputs Outputs Services Information Infrastructure Adaptive Architecture Controller Figure 2.9: Hinnelund's model of an autonomic computing system as a control system. autonomic systems as examples of closed-loop control systems (Taylor and Tofts, 2004). Inputs that cannot be controlled directly are called disturbance inputs. Relationships between inputs and outputs can be linear or non-linear, with linear systems being favoured over non-linear systems due to their predictability. State Monitoring A state-space model can be used to describe a control system that has more than one input and/or output. The size of the state-space should be tractable. The monitoring of system states is equivalent in control systems terms to soliciting feedback from system operation in order to guide the adaptive behaviour of the system (Dutton et al., 1997). State information in control systems usually satis es the Markov property (Dutton et al., 1997), as the decision making components generally make decisions without recourse to historical information (Hinnelund, 2004; Dutton et al., 1997). Techniques have been developed, such as estimators and observers (Dutton et al., 1997), to handle situations where measurements of system state cannot be made directly and must be computed from available measurement data. In Hinnelund's model (see gure 2.9) there is no support for an abstract model of system state. He provides a sample autonomic system in (Hinnelund, 2004) that implements a service work ow. In