Self-adjustment strategy for models used in autonomic transactional systems

This work briefly introduces the architectural design and the operation of an autonomic workload management technique for transactional systems. The technique relies heavily on the capability and knowledge of its manager/controller to predict the behavior of the system with different load conditions. Generally, the manager/controller stores the knowledge about the behavior of the controlled system in form of a model. If the behavior of the system never changes, the model can be estimated and included in the controller only once. But if the behavior of the system varies during its operation, for example due to changes in the implementation of any of the services provided, the model of the system must be readjusted so that it reflects the new behavior of the system properly. The lack of self-adjustment strategies for this type of models has motivated the development of the strategy presented in this paper, devoting special attention to assure that the strategy generates acceptable models during the adjustment period, that is, between the initial and final models.