PANDADB: A Distributed Graph Database System to Query Unstructured Data in Big Graph

Unstructured data, such as images and videos, are growing stringently. The unprecedented growth of interconnected unstructured data can be viewed as graphs, then the node properties of graph can be the unstructured data. End users usually query graph data and unstructured data together in different real-world applications. Some systems and techniques are proposed to meet such demands. However, most of the previous work executes various tasks in different systems and loses the possibility to optimize such queries in one engine. In this work, we build a native graph database namely PandaDB to support querying unstructured data in the graph. We at first introduce CypherPlus, a query language to enable users to express complex graph queries for understanding the semantic of unstructured data. Next, we develop a cost model and related query optimization techniques to speed up the unstructured data processing as well as the graph querying processing. In addition, we optimize the data storage and index to speed up the query processing in a distributed setting. The PandaDB extends the graph database Neo4j implementation and provides the open-source version for commercial use in the cloud. The results show PandaDB can support a large scale of unstructured data query processing in a graph e.g., more than a billion unstructured data items. We also like to share the best practices while deploying the system into real applications.