The technique of Distributed Hash Table (DHT) has become a fundamental part to build peer-to-peer (P2P) overlay network. It provides routing and location algorithms, and assumes each node over the network has equal responsibilities. However, these approaches did not consider the heterogeneous nature and the asymmetric property. Because each node has different properties in computing power, storage space, and network bandwidth, their contributions are frequently asymmetric. This paper presents Japster, an improved approach to P2P networks, which selects a portion of high capability machines that will stay on the network for a long period to be super node for routing accelerator, making all lookup operation finish quickly with three hops. Japster combines DHT and super node to make a good infrastructure with efficiency, self-adaptation and fault-tolerance, scalability, and load-balance.