On the complexity of determining autonomic policy constrained behaviour

Policy Based Management aims to constrain and even to determine the behaviour of computer systems that operate in dynamic environments, e.g. for the implementation of business goals. Autonomic computing supplements this with the aim to allow computer systems operate in a stable and predictable fashion with a minimum of human involvement. In this work we use a formulation of autonomic computing based on the cfengine model of convergent operations to discuss the computational cost of implementing autonomic regulation. By placing the autonomic properties of a system at a low level, but with a high degree of abstraction, we are able to make quite general statements about the computation cost of searching for autonomic policies.