An INLP approach for simulated-based automatic object deployment

Correct deployment of objects over a set of computing resources has a great impact on the performance of a distributed software. In this paper an Integer Non-Linear Programming (INLP) object deployment approach which relies on previous simulation records of the software use-case scenarios is presented. To obtain the optimal deployment of objects, collaborating in a set of use-case scenarios, with respect to response time, the Sequence Diagram corresponding to each scenario is transformed into a set of Finite State Processes (FSP). The resulting FSP's are simulated by means of a discrete event simulator tool to obtain the value of some metrics such as utilization, population and mean response times of objects and threads. These values are good indicators for the execution load of the intended use-case scenarios and along with the resource capacities, are applied to form the objective function of the INLP model. The INLP model then is solved by a solver to determine the optimal object deployment and the best number of replicas corresponding to each active object.