In this paper, we propose a modularized approach for ontology reuse and sharing. In particular, we refer to the context-specific modules from large-scale ontologies as sub-ontologies. We combine the caching mechanism with ontology reuse to form sub-ontology cache. Semantic-based systems are able to take sub-ontology cache as their local knowledge-bases to support semantic-based activities. We also deal with the optimization problem for sub-ontology cache. We propose to optimize the sub-ontologies in a cache with a semantic-based genetic algorithm. Besides, we evaluate the approach with a large-scale traditional Chinese medicine domain ontology to obtain promising results.