Solving the star identification (Star-ID) problem with a rotation-search-based approach eliminates the conventional heuristics in the established paradigms, i.e., the subgraph-isomorphic-based and pattern-recognition-based methods. However, it is not trivial to execute such an approach efficiently. Here, we present ROSIA, which seeks the optimal rotation alignment that maximally matches the input and catalog stars in their respective coordinates. ROSIA searches the rotation space systematically with the Branch-and-Bound (BnB) method. Crucially affecting the runtime feasibility of ROSIA is the upper bound function that prioritizes the search space. In this paper, we make a theoretical contribution by proposing a tight (provable) upper bound function that allows a 400x speed up compared to an existing formulation. Coupling the bounding function with an efficient evaluation scheme that leverages stereographic projection and the R-tree data structure, ROSIA achieves real-time operational speed with state-of-the-art performances under different sources of noise.