Skip to Main Content
The administration of clusters is an exhausting job. Particularly, allocating the resources of the clusters by hand can easily become unmanageable because the processing requirements can change very quickly in a dynamic environment such as the Internet. A solution to solve this problem is to use a dynamic architecture for self-reconfiguration of the clusters. In a previous work, we have proposed the DARC architecture, an agent-based architecture that can perform an automatic reconfiguration to adapt itself to the current needs. In this paper, we formally model our architecture using SPE techniques. These models are validated by comparing the analytical results with results obtained through experimental evaluation. The models obtained can thus be used to evaluate DARC in different environments without the hassle of a time-consuming experimental evaluation. For example, in this paper we have used our models to compare the strategy of load balancing without self-reconfiguration with an approach with self-reconfiguration. This paper also shows that the use of Generalized Stochastic Petri Nets (GSPN) is suitable to analyze complex performance problems in the dynamic reconfiguration domain.