Large graph data in many application domains dynamically changes with vertices and edges inserted and deleted over time. The problem of identifying and maintaining densely connected regions in the graph thus becomes a challenge. Embodiments of the invention describe a method using a k-core measure as a metric of dense connectivity over large, partitioned graph data stored in multiple computing servers in a cluster. The method describes steps to identify a k-core subgraph in parallel and to maintain a k-core subgraph when a new edge is inserted or an existing edge is deleted. The embodiments thus enable practitioners to identify and monitor large scale graph data, such as exemplified by multiple topical communities in a social network, in a scalable and efficient manner.