DHTs are efficient peer-to-peer systems which can locate objects within an bounded amount of overlay hops. Originally, those systems don't exploit network proximity in the underlying Internet and lead to high latency when searching a target. Recently, some approaches, such as Random Landmarking (RLM) and Lookup-Parasitic Random Sampling (LPRS), have been suggested to build topology-aware overlay to improve the lookup efficiency. Although these approaches help the nodes in P2P system to be aware of the underlying network, they do a limited improvement because of their proximity neighbor selection. In this paper, a hashing landmark clusters (HLC) method is proposed to improve the performance of Chord by improving the accuracy of proximity mapping between overlay network and physical network The analysis demonstrates that our approach can reduce the lookup latency better than RLM & LPRS can.