We investigate the problem of modeling the optimal and automatic deployment of cloud applications. We follow an approach based on three main pillars: (i) the specification of the computing resources needed by software components and those provided by the executing environment (e.g. virtual machines or containers), (ii) the declarative description of deployment rules, (iii) and the computation of an optimal deployment that minimizes the total cost by using constraint solving techniques. We experiment with such an approach by applying it to the Abstract Behavioural Specification language ABS, and we validate it by modeling and simulating with ABS (and its tool-suite) the Fredhopper Cloud Services, a worldwide system offering e-Commerce services, currently deployed on Amazon EC2.