Currently, cloud computing systems use simple key-value data processing, which cannot support similarity search effectively due to lack of efficient index structures, and with increase of dimensionality, existing tree-like index structures could lead to problem of the curse of dimensionality. In this paper, a novel VF-CAN indexing scheme is proposed. VF-CAN integrates content addressable network (CAN) based routing protocol and improved vector approximation file (VA-file) index. There are two index levels in this scheme: global index and local index. The local index VAK-file is built for data in each storage node. VAK-file is k-means clustering result of VA-file approximation vectors according to their degree of proximity. Each cluster forms a separate local index file and each file stores approximate vectors that are contained in cluster. The vector of each cluster center is stored in cluster center information file of corresponding storage node. In global index, storage nodes are organized into an overlay network CAN, and in order to reduce cost of calculation, only clustering information of local index is issued to entire overlay network through CAN interface. The experimental results show that VF-CAN reduces index storage space and improves query performance effectively.