Runtime adaptation of service-based software systems (SBS) in cloud environments is a key to acquire resource allocation strategy that meets global optimization goals. However, because of the complex business logic in SBS as well as cloud resource constraints, the optimal resource allocation cannot be obtained using existing methods. This paper offers an approach to meet SLA and minimize resource costs by exploiting the fact that different resource states result in different performance of component services. With the new method, the potential resource allocation for component services, together with responding performance and resource costs, are first transformed into candidate logical service sets. Then a service selection based resource allocation model for SBS in the cloud is constructed. A hybrid genetic algorithm is also designed for solving the model. Integer encoding is applied in the algorithm to improve the efficiency and an elitism maintenance strategy is introduced into selection operator to ensure its convergence to the global optimal solution. In order to improve the local search ability of genetic algorithm and speed up the convergence speed, the standard mutation operator is replaced by local search. Experiments validate the effectiveness of the proposed resource allocation model and its algorithm, and show that the presented algorithm can obtain the resource allocation strategy quickly with lower cost on large-scale problems than the branch and bound method and elitism genetic algorithm.