This paper proposes a domain based adaptive replica selecting model named DARSM (domain based adaptive replica selection model), in which component replicas are organized into strong consistency domain and weak consistency domain, and a schema based on consistency window is utilized to synchronize states between theses domains. Accordingly, a partition-balanced based adaptive replica selection algorithm PWARS (partition- weighted based adaptive replica selection) could be built on dynamic performance metric information to select the appropriate replica set that satisfies specific QoS constraints. A consistency window adaptive reconfiguration algorithm CWAR (consistency window adaptive reconfiguration) is presented to adapt to the dynamic change of consistency constraints. In this algorithm, a probability model built on the base of current consistency constraints distribution is used to dynamically adjust the configuration of consistency window. As a result, the inconsistency of each replica is controlled adaptively. This approach has been implemented in OnceAS, and experimental results demonstrate that it can effectively enhance the performance of replica selection.