Distributed soft policy enforcement by swarm intelligence; application to loadsharing and protectionMise en œuvre de règles par intelligence collective flexible et répartie; application au partage de charge et à la protection

Managing complex heterogeneous computer and telecommunication systems is challenging. One promising management concept for such systems is policy based management. However, it is common to interpret policies strictly and resort to centralized decisions to resolve policy conflicts. Centralization is undesirable from a dependability point of view. Swarm intelligence based on sets of autonomous “ant-like” mobile agents, where control is distribute among the agents, has been applied to several challenging optimization and tradeoff problems with great success. This paper introduces and demonstrates how a set of such ant-like mobile agents can be designed to find near optimal solutions for the implementation of a set of potentially conflicting policies. Solutions are found in a truly distributed manner, hence an overall more dependable/robust system is obtained. The enforcement of the policies is soft in the sense that it is probabilistic and yields a kind of “best effort” implementation. To demonstrate the feasibility of the overall concept, a case study is presented where ant-like mobile agents are designed to implement load distribution and conflict free back-up policies.